Road/Lane Detection Evaluation 2013


This benchmark has been created in collaboration with Jannik Fritsch and Tobias Kuehnl from Honda Research Institute Europe GmbH. The road and lane estimation benchmark consists of 289 training and 290 test images. It contains three different categories of road scenes:

  • uu - urban unmarked (98/100)
  • um - urban marked (95/96)
  • umm - urban multiple marked lanes (96/94)
  • urban - combination of the three above

Ground truth has been generated by manual annotation of the images and is available for two different road terrain types: road - the road area, i.e, the composition of all lanes, and lane - the ego-lane, i.e., the lane the vehicle is currently driving on (only available for category "um"). Ground truth is provided for training images only.

We evaluate road and lane estimation performance in the bird's-eye-view space. For the classical pixel-based evaluation we use established measures as discussed in our ITSC 2013 publication. MaxF: Maximum F1-measure, AP: Average precision as used in PASCAL VOC challenges, PRE: Precision, REC: Recall, FPR: False Positive Rate, FNR: False Negative Rate (the four latter measures are evaluated at the working point MaxF), F1: F1 score, HR: Hit rate. For the novel behavior-based evaluation a corridor with the vehicle width (2.2m) is fitted to the lane estimation processing result and evaluation is performed for 3 different distance values: 20 m, 30 m, and 40 m. We refer to our ITSC 2013 publication for more details.
IMPORTANT NOTE: On 09.02.2015 we have improved the accuracy of the ground truth and re-calculated the results for all methods. Please download the devkit and the dataset with the improved ground truth for training again, if you have downloaded the files prior to 09.02.2015. Please consider reporting these new number for all future submissions. The last leaderboards right before the changes can be found here!

Important Policy Update: As more and more non-published work and re-implementations of existing work is submitted to KITTI, we have established a new policy: from now on, only submissions with significant novelty that are leading to a peer-reviewed paper in a conference or journal are allowed. Minor modifications of existing algorithms or student research projects are not allowed. Such work must be evaluated on a split of the training set. To ensure that our policy is adopted, new users must detail their status, describe their work and specify the targeted venue during registration. Furthermore, we will regularly delete all entries that are 6 months old but are still anonymous or do not have a paper associated with them. For conferences, 6 month is enough to determine if a paper has been accepted and to add the bibliography information. For longer review cycles, you need to resubmit your results.
Additional information used by the methods
  • Stereo: Method uses left and right (stereo) images
  • Laser Points: Method uses point clouds from Velodyne laser scanner
  • GPS: Method uses GPS information
  • Additional training data: Use of additional data sources for training (see details)

Road Estimation Evaluation

UM_ROAD


Method Setting Code MaxF AP PRE REC FPR FNR Runtime Environment
1 DiPFormer 97.29 % 92.66 % 97.37 % 97.20 % 1.20 % 2.80 % 0.01 s 1 GPU @ 2.5 Ghz (C/C++)
S. Chen, T. Han, C. Zhang, W. Liu, J. Su, Z. Wang and G. Cai: Depth Matters: Exploring Deep Interactions of RGB-D for Semantic Segmentation in Traffic Scenes. arXiv preprint arXiv:2409.07995 2024.
2 LiDAR-DHMT 97.29 % 92.66 % 97.37 % 97.20 % 1.20 % 2.80 % 0.01 s 1 core @ 2.5 Ghz (Python)
3 UdeerLID+ 97.26 % 93.54 % 97.38 % 97.15 % 1.19 % 2.85 % 0.01 s 1 core @ 2.5 Ghz (C/C++)
T. Ni, X. Zhan, T. Luo, W. Liu, Z. Shi and J. Chen: UdeerLID+: Integrating LiDAR, Image, and Relative Depth with Semi-Supervised. 2024.
4 SNE-RoadSegV2 97.25 % 93.52 % 97.48 % 97.03 % 1.14 % 2.97 % 0.03 s GPU @ 2.5 Ghz (Python)
Y. Feng, Y. Ma, Q. Chen, I. Pitas and R. Fan: SNE-RoadSegV2: Advancing Heterogeneous Feature Fusion and Fallibility Awareness for Freespace Detection. 2024.
5 RoadFormer+ 97.17 % 93.41 % 97.09 % 97.24 % 1.33 % 2.76 % 0.04 s 1 core @ 2.5 Ghz (C/C++)
J. Huang, J. Li, N. Jia, Y. Sun, C. Liu, Q. Chen and R. Fan: RoadFormer+: Delivering RGB-X Scene Parsing through Scale-Aware Information Decoupling and Advanced Heterogeneous Feature Fusion. IEEE Transactions on Intelligent Vehicles 2024.
6 PLARD
This method makes use of Velodyne laser scans.
code 97.05 % 93.53 % 97.18 % 96.92 % 1.28 % 3.08 % 0.16 s GPU @ 2.5 Ghz (Python)
Z. Chen, J. Zhang and D. Tao: Progressive LiDAR adaptation for road detection. IEEE/CAA Journal of Automatica Sinica 2019.
7 RoadFormer 97.02 % 93.34 % 96.84 % 97.20 % 1.45 % 2.80 % 0.07 s GPU @ 2.5 Ghz (Python)
J. Li, Y. Zhang, P. Yun, G. Zhou, Q. Chen and R. Fan: RoadFormer: Duplex Transformer for RGB-Normal Semantic Road Scene Parsing. IEEE Transactions on Intelligent Vehicles 2024.
8 SNE-RoadSeg+ 96.95 % 93.60 % 96.99 % 96.90 % 1.37 % 3.10 % 0.08 s GPU @ 2.5 Ghz (Python)
H. Wang, R. Fan, P. Cai and M. Liu: SNE-RoadSeg+: Rethinking depth-normal translation and deep supervision for freespace detection. 2021 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) 2021.
9 UdeerLID 96.94 % 93.62 % 97.09 % 96.79 % 1.32 % 3.21 % 0.01 s 1 core @ 2.5 Ghz (C/C++)
T. Ni, X. Zhan, T. Luo, W. Liu, Z. Shi and J. Chen: UdeerLID+: Integrating LiDAR, Image, and Relative Depth with Semi-Supervised. 2024.
10 Pseudo-LiDAR 96.87 % 93.71 % 97.35 % 96.40 % 1.20 % 3.60 % 0.46 s GPU @ 2.5 Ghz (Python)
L. Sun, H. Zhang and W. Yin: Pseudo-LiDAR-Based Road Detection. IEEE Transactions on Circuits and Systems for Video Technology (T-CSVT) 2022.
11 MemRoadNet code 96.55 % 93.58 % 96.46 % 96.64 % 1.61 % 3.36 % 0.01 s GPU @ 2.5 Ghz (Python)
S. Shafiq, A. Khan and J. Shao: MemRoadNet: Human-like Memory Integration for Free Road Space Detection. Sensors 2025.
12 Evi-RoadSeg code 96.51 % 92.94 % 95.90 % 97.13 % 1.89 % 2.87 % 0.01 s GPU @ 2.5 Ghz (Python)
F. Xue✝, Y. Chang✝, W. Xu, W. Liang, F. Sheng and A. Ming: Evidence-based Real-time Road Segmentation with RGB-D Data Augmentation. Transactions on Intelligent Transportation Systems (T-ITS) 2024.
13 USNet code 96.46 % 92.78 % 96.32 % 96.60 % 1.68 % 3.40 % 0.02 s GPU @ 1.5 Ghz (Python)
Y. Chang, F. Xue, F. Sheng, W. Liang and A. Ming: Fast Road Segmentation via Uncertainty-aware Symmetric Network. IEEE International Conference on Robotics and Automation (ICRA) 2022.
14 DFM-RTFNet 96.46 % 93.66 % 96.58 % 96.33 % 1.55 % 3.67 % 0.08 s GPU @ 2.5 Ghz (Python)
H. Wang, R. Fan, Y. Sun and M. Liu: Dynamic fusion module evolves drivable area and road anomaly detection: A benchmark and algorithms. IEEE Transactions on Cybernetics 2021.
15 SNE-RoadSeg code 96.42 % 93.67 % 96.59 % 96.26 % 1.55 % 3.74 % 0.18 s GPU @ 2.5 Ghz (Python)
R. Fan, H. Wang, P. Cai and M. Liu: SNE-RoadSeg: Incorporating Surface Normal Information into Semantic Segmentation for Accurate Freespace Detection. ECCV 2020.
16 3MT-RoadSeg
This method makes use of Velodyne laser scans.
This method makes use of GPS/IMU information.
code 96.13 % 93.42 % 96.20 % 96.06 % 1.73 % 3.94 % 0.07 s GPU @ 2.5 Ghz (Python)
E. Milli, �. Erkent and A. Yılmaz: Multi-Modal Multi-Task (3MT) Road Segmentation. IEEE Robotics and Automation Letters 2023.
17 LRDNet+
This method makes use of Velodyne laser scans.
code 96.10 % 92.00 % 96.89 % 95.32 % 1.39 % 4.68 % 0.01 s GPU @ 2.5 Ghz (Python)
A. Khan, J. Shao, Y. Rao, L. She and H. Shen: LRDNet: Lightweight LiDAR Aided Cascaded Feature Pools for Free Road Space Detection. IEEE Transactions on Multimedia 2025.
18 LRDNet(S)
This method makes use of Velodyne laser scans.
code 96.01 % 92.47 % 96.60 % 95.43 % 1.53 % 4.57 % .009 s GPU @ 2.5 Ghz (Python)
A. Khan, J. Shao, Y. Rao, L. She and H. Shen: LRDNet: Lightweight LiDAR Aided Cascaded Feature Pools for Free Road Space Detection. IEEE Transactions on Multimedia 2025.
19 LRDNet (L)
This method makes use of Velodyne laser scans.
code 96.01 % 91.83 % 96.84 % 95.19 % 1.41 % 4.81 % 0.1 s GPU @ 2.5 Ghz (Python)
A. Khan, J. Shao, Y. Rao, L. She and H. Shen: LRDNet: Lightweight LiDAR Aided Cascaded Feature Pools for Free Road Space Detection. IEEE Transactions on Multimedia 2025.
20 RBANet 95.78 % 89.14 % 94.92 % 96.66 % 2.36 % 3.34 % 0.16 s GPU @ 1.5 Ghz (Python + C/C++)
J. Sun, S. Kim, S. Lee, Y. Kim and S. Ko: Reverse and Boundary Attention Network for Road Segmentation. Proceedings of the IEEE International Conference on Computer Vision Workshops 2019.
21 NIM-RTFNet 95.71 % 93.56 % 95.84 % 95.59 % 1.89 % 4.41 % 0.05 s GPU @ 2.5 Ghz (Python)
H. Wang, R. Fan, Y. Sun and M. Liu: Applying Surface Normal Information in Drivable Area and Road Anomaly Detection for Ground Mobile Robots. 2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) 2020.
22 CLCFNet
This method makes use of Velodyne laser scans.
95.65 % 89.49 % 95.31 % 96.00 % 2.15 % 4.00 % 0.02 s GPU @ 1.5 Ghz (Python)
S. Gu, J. Yang and H. Kong: A Cascaded LiDAR-Camera Fusion Network for Road Detection. ICRA 2021.
23 LidCamNet
This method makes use of Velodyne laser scans.
95.62 % 93.54 % 95.77 % 95.48 % 1.92 % 4.52 % 0.15 s GPU @ 2.5 Ghz (Python)
L. Caltagirone, M. Bellone, L. Svensson and M. Wahde: LIDAR-Camera Fusion for Road Detection Using Fully Convolutional Neural Networks. Robotics and Autonomous Systems 2018.
24 CLCFNet (LiDAR)
This method makes use of Velodyne laser scans.
95.16 % 89.18 % 94.97 % 95.36 % 2.30 % 4.64 % 0.02 s GPU @ 1.5 Ghz (C/C++)
S. Gu, J. Yang and H. Kong: A Cascaded LiDAR-Camera Fusion Network for Road Detection. ICRA 2021.
25 TVFNet
This method makes use of Velodyne laser scans.
94.96 % 89.17 % 94.95 % 94.97 % 2.30 % 5.03 % 0.04 s GPU @ 1.5 Ghz (Python)
S. Gu, Y. Zhang, J. Yang, J. Alvarez and H. Kong: Two-View Fusion based Convolutional Neural Network for Urban Road Detection. IROS 2019.
26 LC-CRF
This method makes use of Velodyne laser scans.
94.91 % 86.41 % 91.92 % 98.11 % 3.93 % 1.89 % 0.18 s GPU @ 1.5 Ghz (Python + C/C++)
S. Gu, Y. Zhang, J. Tang, J. Yang and H. Kong: Road Detection through CRF based LiDAR-Camera Fusion. ICRA 2019.
27 RBNet 94.77 % 91.42 % 95.16 % 94.37 % 2.19 % 5.63 % 0.18 s GPU @ 2.5 Ghz (Matlab + C/C++)
Z. Chen and Z. Chen: RBNet: A Deep Neural Network for Unified Road and Road Boundary Detection. International Conference on Neural Information Processing 2017.
28 SSLGAN 94.62 % 89.50 % 95.32 % 93.93 % 2.10 % 6.07 % 700ms GPU @ 1.5 Ghz (Python)
X. Han, J. Lu, C. Zhao, S. You and H. Li: Semisupervised and Weakly Supervised Road Detection Based on Generative Adversarial Networks. IEEE Signal Processing Letters 2018.
29 LFD-RoadSeg code 94.58 % 93.42 % 95.20 % 93.98 % 2.16 % 6.02 % .004 s GPU @ 1.5 Ghz (Python)
H. Zhou, F. Xue, Y. Li, S. Gong, Y. Li and Y. Zhou: Exploiting Low-Level Representations for Ultra-Fast Road Segmentation. IEEE Transactions on Intelligent Transportation Systems 2024.
30 RGB36-Cotrain 94.55 % 93.12 % 94.81 % 94.29 % 2.35 % 5.71 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
L. Caltagirone, S. Lennart, M. Wahde and M. Sanfridson: Lidar-Camera Co-Training for Semi- Supervised Road Detection. arXiv preprint arXiv:1911.12597 2019.
31 HA-DeepLabv3+ 94.38 % 92.72 % 94.70 % 94.06 % 2.40 % 5.94 % 0.06 s 1 core @ 2.5 Ghz (Python)
R. Fan, H. Wang, P. Cai, J. Wu, M. Bocus, L. Qiao and M. Liu: Learning collision-free space detection from stereo images: Homography matrix brings better data augmentation. IEEE/ASME Transactions on Mechatronics 2021.
32 TEDNet
This method makes use of Velodyne laser scans.
code 94.24 % 92.43 % 93.45 % 95.04 % 3.04 % 4.96 % 0.09 s GPU @ 2.5 Ghz (Python)
M. Bayón-Gutiérrez, M. García- Ordás, H. Alaiz Moretón, J. Aveleira-Mata, S. Rubio-Martín and J. Benítez-Andrades: TEDNet: Twin Encoder Decoder Neural Network for 2D Camera and LiDAR Road Detection. Logic Journal of the IGPL 2024.
33 BJN 94.17 % 89.16 % 94.95 % 93.41 % 2.26 % 6.59 % 0.02 s 1 core @ 2.5 Ghz (C/C++)
B. Yu, D. Lee, J. Lee and S. Kee: Free Space Detection Using Camera-LiDAR Fusion in a Bird’s Eye View Plane. Sensors 2021.
34 DEEP-DIG 94.16 % 93.41 % 95.02 % 93.32 % 2.23 % 6.68 % 0.14 s GPU @ 3.5 Ghz (Python + C/C++)
J. Muñoz-Bulnes, C. Fernandez, I. Parra, D. Fernández-Llorca and M. Sotelo: Deep Fully Convolutional Networks with Random Data Augmentation for Enhanced Generalization in Road Detection. Workshop on Deep Learning for Autonomous Driving on IEEE 20th International Conference on Intelligent Transportation Systems 2017.
35 CLRD
This method makes use of Velodyne laser scans.
94.06 % 92.13 % 94.32 % 93.80 % 2.57 % 6.21 % 0.05 s GPU @ 2.5 Ghz (Python)
M. Bayón-Gutiérrez, J. Benítez- Andrades, S. Rubio-Martín, J. Aveleira-Mata, H. Alaiz-Moretón and M. García-Ordás: Roadway Detection Using Convolutional Neural Network Through Camera and LiDAR Data. Hybrid Artificial Intelligent Systems 2022.
36 Hadamard-FCN 94.06 % 90.89 % 94.62 % 93.50 % 2.42 % 6.50 % 0.02 s GPU @ 1.5 Ghz (Python)
M. Oeljeklaus: An Integrated Approach for Traffic Scene Understanding from Monocular Cameras. 2021.
37 StixelNet II 94.05 % 85.85 % 91.30 % 96.98 % 4.21 % 3.02 % 1.2 s 1 core @ 3.0 Ghz (Matlab + C/C++)
N. Garnett, S. Silberstein, S. Oron, E. Fetaya, U. Verner, A. Ayash, V. Goldner, R. Cohen, K. Horn and D. Levi: Real-time category-based and general obstacle detection for autonomous driving. 5th Workshop on Computer Vision for Road Scene Understanding and Autonomous Driving (CVRSUAD'17, IEEE-ICCV 2017 Workshop) 2017.
38 MultiNet code 93.99 % 93.24 % 94.51 % 93.48 % 2.47 % 6.52 % 0.17 s GPU @ 2.5 Ghz (Python + C/C++)
M. Teichmann, M. Weber, J. Zoellner, R. Cipolla and R. Urtasun: MultiNet: Real-time Joint Semantic Reasoning for Autonomous Driving. CoRR 2016.
39 ChipNet
This method makes use of Velodyne laser scans.
93.73 % 87.62 % 93.25 % 94.21 % 3.11 % 5.79 % 12 ms GPU @ 1.5 Ghz (Keras)
Y. Lyu, L. Bai and X. Huang: ChipNet: Real-Time LiDAR Processing for Drivable Region Segmentation on an FPGA. IEEE Transactions on Circuits and Systems I: Regular Papers 2019.
40 DDN 93.65 % 88.55 % 94.28 % 93.03 % 2.57 % 6.97 % 2 s GPU @ 2.5 Ghz (Python + C/C++)
R. Mohan: Deep Deconvolutional Networks for Scene Parsing. 2014.
41 RoadNet3 93.54 % 92.64 % 93.65 % 93.44 % 2.89 % 6.56 % 300 ms GPU @ GTX950M (Python +Tensorflow)
Y. Lyu, L. Bai and X. Huang: Road Segmentation using CNN and Distributed LSTM. 2019 IEEE International Symposium on Circuits and Systems (ISCAS) 2019.
42 HID-LS
This method makes use of Velodyne laser scans.
93.10 % 86.38 % 91.89 % 94.33 % 3.79 % 5.67 % 0.25 s 1 cores @ 3.0 Ghz (C/C++)
S. Gu, Y. Zhang, J. Yang and H. Kong: Lidar-based urban road detection by histograms of normalized inverse depths and line scanning. ECMR 2017.
S. Gu, Y. Zhang, X. Yuan, J. Yang, T. Wu and H. Kong: Histograms of the Normalized Inverse Depth and Line Scanning for Urban Road Detection. IEEE Trans. Intelligent Transportation Systems 2019.
43 RGB36-Super 93.04 % 91.85 % 93.62 % 92.46 % 2.87 % 7.54 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
L. Caltagirone, S. Lennart, M. Wahde and M. Sanfridson: Lidar-Camera Co-Training for Semi- Supervised Road Detection. arXiv preprint arXiv:1911.12597 2019.
44 LoDNN
This method makes use of Velodyne laser scans.
92.75 % 89.98 % 90.09 % 95.58 % 4.79 % 4.42 % 18 ms GPU @ 2.5 Ghz (Torch)
L. Caltagirone, S. Scheidegger, L. Svensson and M. Wahde: Fast LIDAR-based Road Detection Using Fully Convolutional Neural Networks. IEEE Intelligent Vehicles Symposium 2017.
45 Up-Conv-Poly code 92.20 % 88.85 % 92.57 % 91.83 % 3.36 % 8.17 % 0.08 s GPU @ 2.5 Ghz (Python + C/C++)
G. Oliveira, W. Burgard and T. Brox: Efficient Deep Methods for Monocular Road Segmentation. IROS 2016.
46 OFA Net code 92.08 % 82.73 % 87.87 % 96.72 % 6.08 % 3.28 % 0.04 s GPU @ 1.5 Ghz (Python)
S. Zhang, Z. Zhang, L. Sun and W. Qin: One For All: A Mutual Enhancement Method for Object Detection and Semantic Segmentation. Applied Sciences 2019.
47 RoadNet-RT 91.99 % 92.54 % 92.75 % 91.24 % 3.25 % 8.76 % 8m s GPU @ 2.5 Ghz (Python)
L. Bai, Y. Lyu and X. Huang: RoadNet-RT: High Throughput CNN Architecture and SoC Design for Real-Time Road Segmentation. arXiv preprint arXiv:2006.07644 2020.
48 MixedCRF
This method makes use of Velodyne laser scans.
91.57 % 84.68 % 90.02 % 93.19 % 4.71 % 6.81 % 6s 1 core @ 2.5 Ghz (Matlab + C/C++)
X. Han, H. Wang, J. Lu and C. Zhao: Road detection based on the fusion of Lidar and image data. 2017.
49 FTP 91.20 % 90.60 % 91.11 % 91.29 % 4.06 % 8.71 % 0.28 s GPU @ 2.5 Ghz (C/C++)
A. Laddha, M. Kocamaz, L. Navarro-Serment and M. Hebert: Map-Supervised Road Detection. IEEE Intelligent Vehicles Symposium Proceedings 2016.
50 ALO-AVG-MM code 91.15 % 83.82 % 89.07 % 93.33 % 5.22 % 6.67 % 0.0296 sec GeForce GTX 1080 GPU (Python)
F. Reis, R. Almeida, E. Kijak, S. Malinowski, S. Guimaraes and Z. Jr.: Combining convolutional side-outputs for road image segmentation. 2019 International Joint Conference on Neural Networks (IJCNN) - \textbfAccepted 2019.
51 HybridCRF
This method makes use of Velodyne laser scans.
90.99 % 85.26 % 90.65 % 91.33 % 4.29 % 8.67 % 1.5 s 1 core @ 2.5 Ghz (C/C++)
L. Xiao, R. Wang, B. Dai, Y. Fang, D. Liu and T. Wu: Hybrid conditional random field based camera-LIDAR fusion for road detection. Information Sciences 2018.
52 NNP
This method uses stereo information.
90.50 % 87.95 % 91.43 % 89.59 % 3.83 % 10.41 % 5 s 4 cores @ 2.5 Ghz (Matlab)
X. Chen, K. Kundu, Y. Zhu, A. Berneshawi, H. Ma, S. Fidler and R. Urtasun: 3D Object Proposals for Accurate Object Class Detection. NIPS 2015.
53 Up-Conv 90.48 % 88.20 % 91.30 % 89.67 % 3.90 % 10.33 % 0.05 s GPU @ 2.5 Ghz (C/C++)
G. Oliveira, W. Burgard and T. Brox: Efficient Deep Methods for Monocular Road Segmentation. IROS 2016.
54 HIM 90.07 % 79.98 % 90.79 % 89.35 % 4.13 % 10.65 % 7 s >8 cores @ 2.5 Ghz (Python + C/C++)
D. Munoz, J. Bagnell and M. Hebert: Stacked Hierarchical Labeling. European Conference on Computer Vision (ECCV) 2010.
55 LidarHisto
This method makes use of Velodyne laser scans.
code 89.87 % 83.03 % 91.28 % 88.49 % 3.85 % 11.51 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
L. Chen, J. Yang and H. Kong: Lidar-histogram for fast road and obstacle detection. 2017 IEEE International Conference on Robotics and Automation (ICRA) 2017.
56 FusedCRF
This method makes use of Velodyne laser scans.
89.55 % 80.00 % 84.87 % 94.78 % 7.70 % 5.22 % 2 s 1 core @ 2.5 Ghz (C/C++)
L. Xiao, B. Dai, D. Liu, T. Hu and T. Wu: CRF based Road Detection with Multi-Sensor Fusion. Intelligent Vehicles Symposium (IV) 2015.
57 BMCF
This method uses stereo information.
89.42 % 83.13 % 88.31 % 90.55 % 5.46 % 9.45 % 2.5 s 1 core @ 2.5 Ghz (C/C++)
L. Wang, T. Wu, Z. Xiao, L. Xiao, D. Zhao and J. Han: Multi-cue road boundary detection using stereo vision. 2016 IEEE International Conference on Vehicular Electronics and Safety (ICVES) 2016.
58 FCN-LC 89.36 % 78.80 % 89.35 % 89.37 % 4.85 % 10.63 % 0.03 s GPU Titan X
C. Mendes, V. Frémont and D. Wolf: Exploiting Fully Convolutional Neural Networks for Fast Road Detection. IEEE Conference on Robotics and Automation (ICRA) 2016.
59 CB 88.89 % 82.17 % 87.26 % 90.58 % 6.03 % 9.42 % 2 s 1 core @ 3.4 Ghz (Python) + GPU
C. Mendes, V. Frémont and D. Wolf: Vision-Based Road Detection using Contextual Blocks. 2015.
60 SPRAY 88.14 % 91.24 % 88.60 % 87.68 % 5.14 % 12.32 % 45 ms NVIDIA GTX 580 (Python + OpenCL)
T. Kuehnl, F. Kummert and J. Fritsch: Spatial Ray Features for Real-Time Ego-Lane Extraction. Proc. IEEE Intelligent Transportation Systems 2012.
61 ProbBoost
This method uses stereo information.
87.48 % 80.13 % 85.02 % 90.09 % 7.23 % 9.91 % 2.5 min >8 cores @ 3.0 Ghz (C/C++)
G. Vitor, A. Victorino and J. Ferreira: A probabilistic distribution approach for the classification of urban roads in complex environments. Workshop on Modelling, Estimation, Perception and Control of All Terrain Mobile Robots on IEEE International Conference on Robotics and Automation (ICRA) 2014.
62 MAP 87.33 % 89.62 % 85.77 % 88.95 % 6.73 % 11.05 % 0.28s GPU
A. Laddha, M. Kocamaz, L. Navarro-Serment and M. Hebert: Map-Supervised Road Detection. IEEE Intelligent Vehicles Symposium Proceedings 2016.
63 CN24 86.32 % 89.19 % 87.80 % 84.89 % 5.37 % 15.11 % 30 s >8 cores @ 2.5 Ghz (C/C++)
C. Brust, S. Sickert, M. Simon, E. Rodner and J. Denzler: Convolutional Patch Networks with Spatial Prior for Road Detection and Urban Scene Understanding. VISAPP 2015 - Proceedings of the 10th International Conference on Computer Vision Theory and Applications, Berlin, Germany, 11-14 March, 2015 2015.
64 multi-task CNN 85.95 % 81.28 % 77.40 % 96.64 % 12.86 % 3.36 % 25.1 ms GPU @ 2.0 Ghz (Python)
M. Oeljeklaus, F. Hoffmann and T. Bertram: A Fast Multi-Task CNN for Spatial Understanding of Traffic Scenes. IEEE Intelligent Transportation Systems Conference 2018.
65 GRES3D+VELO
This method makes use of Velodyne laser scans.
85.43 % 83.04 % 82.69 % 88.37 % 8.43 % 11.63 % 60 ms 4 cores @ 2.8 Ghz (C/C++)
P. Shinzato: Estimation of obstacles and road area with sparse 3D points. 2015.
66 StixelNet 85.33 % 72.14 % 81.21 % 89.89 % 9.48 % 10.11 % 1 s GPU @ 2.5 Ghz (C/C++)
D. Levi, N. Garnett and E. Fetaya: StixelNet: A Deep Convolutional Network for Obstacle Detection and Road Segmentation.. 26TH British Machine Vision Conference (BMVC) 2015.
67 SPlane + BL
This method uses stereo information.
85.23 % 88.66 % 83.43 % 87.12 % 7.89 % 12.88 % 2 s 1 core @ 3.0 Ghz (C/C++)
N. Einecke and J. Eggert: Block-Matching Stereo with Relaxed Fronto-Parallel Assumption. IV 2014.
68 geo+gpr+crf
This method uses stereo information.
85.13 % 72.24 % 81.33 % 89.29 % 9.34 % 10.71 % 30 s 1 core @ 2.0 Ghz (C/C++)
Z. Xiao, B. Dai, H. Li, T. Wu, X. Xu, Y. Zeng and T. Chen: Gaussian process regression-based robust free space detection for autonomous vehicle by 3-D point cloud and 2-D appearance information fusion. International Journal of Advanced Robotic Systems 2017.
69 RES3D-Velo
This method makes use of Velodyne laser scans.
83.81 % 73.95 % 78.56 % 89.80 % 11.16 % 10.20 % 0.36 s 1 core @ 2.5 Ghz (C/C++)
P. Shinzato, D. Wolf and C. Stiller: Road Terrain Detection: Avoiding Common Obstacle Detection Assumptions Using Sensor Fusion. Intelligent Vehicles Symposium (IV) 2014.
70 SCRFFPFHGSP
This method uses stereo information.
83.73 % 72.89 % 82.13 % 85.39 % 8.47 % 14.61 % 5 s 8 cores @ 2.5 Ghz (C/C++, Matlab)
I. Gheorghe: Semantic Segmentation of Terrain and Road Terrain for Advanced Driver Assistance Systems. 2015.
71 GRES3D+SELAS
This method uses stereo information.
83.69 % 84.61 % 78.31 % 89.88 % 11.35 % 10.12 % 110 ms 4 core @ 2.8 Ghz (C/C++)
P. Shinzato: Estimation of obstacles and road area with sparse 3D points. 2015.
72 HistonBoost
This method uses stereo information.
83.68 % 72.79 % 82.01 % 85.42 % 8.54 % 14.58 % 2.5 min >8 cores @ 3.0 Ghz (C/C++)
G. Vitor, A. Victorino and J. Ferreira: Comprehensive Performance Analysis of Road Detection Algorithms Using the Common Urban Kitti-Road Benchmark. Workshop on Benchmarking Road Terrain and Lane Detection Algorithms for In-Vehicle Application on IEEE Intelligent Vehicles Symposium (IV) 2014.
73 PGM-ARS 80.97 % 69.11 % 77.51 % 84.76 % 11.21 % 15.24 % 0.05 s i74700MQ @ 2.1Ghz (C/C++)
M. Passani, J. Yebes and L. Bergasa: Fast Pixelwise Road Inference Based on Uniformly Reweighted Belief Propagation . Proc. IEEE Intelligent Vehicles Symposium 2015.
74 RES3D-Stereo
This method uses stereo information.
78.98 % 80.06 % 75.94 % 82.27 % 11.88 % 17.73 % 0.7 s 1 core @ 2.5 Ghz (C/C++)
P. Shinzato, D. Gomes and D. Wolf: Road estimation with sparse 3D points from stereo data. Intelligent Transportation Systems (ITSC), 2014 IEEE 17th International Conference on 2014.
75 BM
This method uses stereo information.
78.90 % 66.06 % 69.53 % 91.19 % 18.21 % 8.81 % 2 s 2 cores @ 2.5 Ghz (Matlab)
B. WANG, V. Fremont and S. Rodriguez Florez: Color-based Road Detection and its Evaluation on the KITTI Road Benchmark. Workshop on Benchmarking Road Terrain and Lane Detection Algorithms for In-Vehicle Application, IEEE Intelligent Vehicles Symposium 2014.
76 SPlane
This method uses stereo information.
78.19 % 76.41 % 72.03 % 85.50 % 15.13 % 14.50 % 2 s 1 core @ 3.0 Ghz (C/C++)
N. Einecke and J. Eggert: Block-Matching Stereo with Relaxed Fronto-Parallel Assumption. IV 2014.
77 SRF 76.43 % 83.24 % 75.53 % 77.35 % 11.42 % 22.65 % 0.2 s 1 core @ 2.5 Ghz (C/C++)
L. Xiao, B. Dai, D. Liu, D. Zhao and T. Wu: Monocular Road Detection Using Structured Random Forest. Int J Adv Robot Syst 2016.
78 CN24 76.28 % 79.29 % 72.44 % 80.55 % 13.96 % 19.45 % 20 s >8 cores @ 2.5 Ghz (C/C++)
C. Brust, S. Sickert, M. Simon, E. Rodner and J. Denzler: Convolutional Patch Networks with Spatial Prior for Road Detection and Urban Scene Understanding. VISAPP 2015 - Proceedings of the 10th International Conference on Computer Vision Theory and Applications, Berlin, Germany, 11-14 March, 2015 2015.
79 CN 73.69 % 76.68 % 69.18 % 78.83 % 16.00 % 21.17 % 2 s 1 core @ 2.5 Ghz (C/C++)
J. Alvarez, T. Gevers, Y. LeCun and A. Lopez: Road Scene Segmentation from a Single Image. ECCV 2012 2012.
80 ARSL-AMI 71.97 % 61.04 % 78.03 % 66.79 % 8.57 % 33.21 % 0.05 s 4 cores @ 2.5 Ghz (C/C++)
M. Passani, J. Yebes and L. Bergasa: CRF-based semantic labeling in miniaturized road scenes . Proc. IEEE Intelligent Transportation Systems 2014.
81 ANN
This method uses stereo information.
62.83 % 46.77 % 50.21 % 83.91 % 37.91 % 16.09 % 3 s 1 core @ 3.0 Ghz (C/C++)
G. Vitor, D. Lima, A. Victorino and J. Ferreira: A 2D/3D Vision Based Approach Applied to Road Detection in Urban Environments. Intelligent Vehicles Symposium (IV), 2013 IEEE 2013.
Table as LaTeX | Only published Methods

UMM_ROAD


Method Setting Code MaxF AP PRE REC FPR FNR Runtime Environment
1 DiPFormer 98.19 % 94.80 % 97.88 % 98.50 % 2.34 % 1.50 % 0.01 s 1 GPU @ 2.5 Ghz (C/C++)
S. Chen, T. Han, C. Zhang, W. Liu, J. Su, Z. Wang and G. Cai: Depth Matters: Exploring Deep Interactions of RGB-D for Semantic Segmentation in Traffic Scenes. arXiv preprint arXiv:2409.07995 2024.
2 LiDAR-DHMT 98.19 % 94.80 % 97.88 % 98.50 % 2.34 % 1.50 % 0.01 s 1 core @ 2.5 Ghz (Python)
3 RoadFormer+ 98.15 % 95.46 % 98.07 % 98.23 % 2.12 % 1.77 % 0.04 s 1 core @ 2.5 Ghz (C/C++)
J. Huang, J. Li, N. Jia, Y. Sun, C. Liu, Q. Chen and R. Fan: RoadFormer+: Delivering RGB-X Scene Parsing through Scale-Aware Information Decoupling and Advanced Heterogeneous Feature Fusion. IEEE Transactions on Intelligent Vehicles 2024.
4 RoadFormer 98.15 % 95.60 % 98.07 % 98.23 % 2.13 % 1.77 % 0.07 s GPU @ 2.5 Ghz (Python)
J. Li, Y. Zhang, P. Yun, G. Zhou, Q. Chen and R. Fan: RoadFormer: Duplex Transformer for RGB-Normal Semantic Road Scene Parsing. IEEE Transactions on Intelligent Vehicles 2024.
5 SNE-RoadSeg+ 98.13 % 95.52 % 98.01 % 98.25 % 2.19 % 1.75 % 0.08 s GPU @ 2.5 Ghz (Python)
H. Wang, R. Fan, P. Cai and M. Liu: SNE-RoadSeg+: Rethinking depth-normal translation and deep supervision for freespace detection. 2021 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) 2021.
6 SNE-RoadSegV2 98.10 % 95.63 % 97.98 % 98.22 % 2.23 % 1.78 % 0.03 s GPU @ 2.5 Ghz (Python)
Y. Feng, Y. Ma, Q. Chen, I. Pitas and R. Fan: SNE-RoadSegV2: Advancing Heterogeneous Feature Fusion and Fallibility Awareness for Freespace Detection. 2024.
7 Pseudo-LiDAR 98.05 % 95.63 % 97.89 % 98.21 % 2.33 % 1.79 % 0.46 s GPU @ 2.5 Ghz (Python)
L. Sun, H. Zhang and W. Yin: Pseudo-LiDAR-Based Road Detection. IEEE Transactions on Circuits and Systems for Video Technology (T-CSVT) 2022.
8 UdeerLID+ 98.02 % 95.50 % 97.80 % 98.25 % 2.43 % 1.75 % 0.01 s 1 core @ 2.5 Ghz (C/C++)
T. Ni, X. Zhan, T. Luo, W. Liu, Z. Shi and J. Chen: UdeerLID+: Integrating LiDAR, Image, and Relative Depth with Semi-Supervised. 2024.
9 LRDNet+
This method makes use of Velodyne laser scans.
code 97.98 % 94.28 % 97.67 % 98.29 % 2.58 % 1.71 % 0.01 s GPU @ 2.5 Ghz (Python)
A. Khan, J. Shao, Y. Rao, L. She and H. Shen: LRDNet: Lightweight LiDAR Aided Cascaded Feature Pools for Free Road Space Detection. IEEE Transactions on Multimedia 2025.
10 LRDNet (L)
This method makes use of Velodyne laser scans.
code 97.91 % 93.96 % 97.45 % 98.37 % 2.83 % 1.63 % 0.1 s GPU @ 2.5 Ghz (Python)
A. Khan, J. Shao, Y. Rao, L. She and H. Shen: LRDNet: Lightweight LiDAR Aided Cascaded Feature Pools for Free Road Space Detection. IEEE Transactions on Multimedia 2025.
11 UdeerLID 97.90 % 95.58 % 97.64 % 98.17 % 2.61 % 1.83 % 0.01 s 1 core @ 2.5 Ghz (C/C++)
T. Ni, X. Zhan, T. Luo, W. Liu, Z. Shi and J. Chen: UdeerLID+: Integrating LiDAR, Image, and Relative Depth with Semi-Supervised. 2024.
12 LRDNet(S)
This method makes use of Velodyne laser scans.
code 97.82 % 94.29 % 97.39 % 98.25 % 2.90 % 1.75 % .009 s GPU @ 2.5 Ghz (Python)
A. Khan, J. Shao, Y. Rao, L. She and H. Shen: LRDNet: Lightweight LiDAR Aided Cascaded Feature Pools for Free Road Space Detection. IEEE Transactions on Multimedia 2025.
13 Evi-RoadSeg code 97.80 % 95.25 % 97.50 % 98.10 % 2.77 % 1.90 % 0.01 s GPU @ 2.5 Ghz (Python)
F. Xue✝, Y. Chang✝, W. Xu, W. Liang, F. Sheng and A. Ming: Evidence-based Real-time Road Segmentation with RGB-D Data Augmentation. Transactions on Intelligent Transportation Systems (T-ITS) 2024.
14 PLARD
This method makes use of Velodyne laser scans.
code 97.77 % 95.64 % 97.75 % 97.79 % 2.48 % 2.21 % 0.16 s GPU @ 2.5 Ghz (Python)
Z. Chen, J. Zhang and D. Tao: Progressive LiDAR adaptation for road detection. IEEE/CAA Journal of Automatica Sinica 2019.
15 USNet code 97.68 % 95.13 % 97.28 % 98.09 % 3.02 % 1.91 % 0.02 s GPU @ 1.5 Ghz (Python)
Y. Chang, F. Xue, F. Sheng, W. Liang and A. Ming: Fast Road Segmentation via Uncertainty-aware Symmetric Network. IEEE International Conference on Robotics and Automation (ICRA) 2022.
16 SNE-RoadSeg code 97.47 % 95.63 % 97.32 % 97.61 % 2.96 % 2.39 % 0.18 s GPU @ 2.5 Ghz (Python)
R. Fan, H. Wang, P. Cai and M. Liu: SNE-RoadSeg: Incorporating Surface Normal Information into Semantic Segmentation for Accurate Freespace Detection. ECCV 2020.
17 MemRoadNet code 97.46 % 95.57 % 97.07 % 97.86 % 3.25 % 2.14 % 0.01 s GPU @ 2.5 Ghz (Python)
S. Shafiq, A. Khan and J. Shao: MemRoadNet: Human-like Memory Integration for Free Road Space Detection. Sensors 2025.
18 3MT-RoadSeg
This method makes use of Velodyne laser scans.
This method makes use of GPS/IMU information.
code 97.46 % 95.54 % 97.30 % 97.62 % 2.97 % 2.38 % 0.07 s GPU @ 2.5 Ghz (Python)
E. Milli, �. Erkent and A. Yılmaz: Multi-Modal Multi-Task (3MT) Road Segmentation. IEEE Robotics and Automation Letters 2023.
19 DFM-RTFNet 97.45 % 95.63 % 97.33 % 97.58 % 2.94 % 2.42 % 0.08 s GPU @ 2.5 Ghz (Python)
H. Wang, R. Fan, Y. Sun and M. Liu: Dynamic fusion module evolves drivable area and road anomaly detection: A benchmark and algorithms. IEEE Transactions on Cybernetics 2021.
20 RBANet 97.38 % 92.67 % 96.70 % 98.08 % 3.68 % 1.92 % 0.16 s GPU @ 1.5 Ghz (Python + C/C++)
J. Sun, S. Kim, S. Lee, Y. Kim and S. Ko: Reverse and Boundary Attention Network for Road Segmentation. Proceedings of the IEEE International Conference on Computer Vision Workshops 2019.
21 CLCFNet
This method makes use of Velodyne laser scans.
97.24 % 93.84 % 97.99 % 96.51 % 2.18 % 3.49 % 0.02 s GPU @ 1.5 Ghz (Python)
S. Gu, J. Yang and H. Kong: A Cascaded LiDAR-Camera Fusion Network for Road Detection. ICRA 2021.
22 LC-CRF
This method makes use of Velodyne laser scans.
97.08 % 92.06 % 96.03 % 98.16 % 4.46 % 1.84 % 0.18 s GPU @ 1.5 Ghz (Python + C/C++)
S. Gu, Y. Zhang, J. Tang, J. Yang and H. Kong: Road Detection through CRF based LiDAR-Camera Fusion. ICRA 2019.
23 LidCamNet
This method makes use of Velodyne laser scans.
97.08 % 95.51 % 97.28 % 96.88 % 2.98 % 3.12 % 0.15 s GPU @ 2.5 Ghz (Python)
L. Caltagirone, M. Bellone, L. Svensson and M. Wahde: LIDAR-Camera Fusion for Road Detection Using Fully Convolutional Neural Networks. Robotics and Autonomous Systems 2018.
24 CLCFNet (LiDAR)
This method makes use of Velodyne laser scans.
96.88 % 93.71 % 97.85 % 95.94 % 2.32 % 4.06 % 0.02 s GPU @ 1.5 Ghz (C/C++)
S. Gu, J. Yang and H. Kong: A Cascaded LiDAR-Camera Fusion Network for Road Detection. ICRA 2021.
25 NIM-RTFNet 96.79 % 95.61 % 97.03 % 96.54 % 3.25 % 3.46 % 0.05 s GPU @ 2.5 Ghz (Python)
H. Wang, R. Fan, Y. Sun and M. Liu: Applying Surface Normal Information in Drivable Area and Road Anomaly Detection for Ground Mobile Robots. 2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) 2020.
26 RGB36-Cotrain 96.75 % 95.39 % 96.84 % 96.66 % 3.46 % 3.34 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
L. Caltagirone, S. Lennart, M. Wahde and M. Sanfridson: Lidar-Camera Co-Training for Semi- Supervised Road Detection. arXiv preprint arXiv:1911.12597 2019.
27 SSLGAN 96.72 % 92.99 % 97.05 % 96.40 % 3.22 % 3.60 % 700ms GPU @ 1.5 Ghz (Python)
X. Han, J. Lu, C. Zhao, S. You and H. Li: Semisupervised and Weakly Supervised Road Detection Based on Generative Adversarial Networks. IEEE Signal Processing Letters 2018.
28 LFD-RoadSeg code 96.59 % 95.40 % 96.29 % 96.90 % 4.11 % 3.10 % .004 s GPU @ 1.5 Ghz (Python)
H. Zhou, F. Xue, Y. Li, S. Gong, Y. Li and Y. Zhou: Exploiting Low-Level Representations for Ultra-Fast Road Segmentation. IEEE Transactions on Intelligent Transportation Systems 2024.
29 TVFNet
This method makes use of Velodyne laser scans.
96.47 % 93.16 % 97.24 % 95.71 % 2.98 % 4.29 % 0.04 s GPU @ 1.5 Ghz (Python)
S. Gu, Y. Zhang, J. Yang, J. Alvarez and H. Kong: Two-View Fusion based Convolutional Neural Network for Urban Road Detection. IROS 2019.
30 BJN 96.29 % 93.98 % 98.14 % 94.52 % 1.97 % 5.48 % 0.02 s 1 core @ 2.5 Ghz (C/C++)
B. Yu, D. Lee, J. Lee and S. Kee: Free Space Detection Using Camera-LiDAR Fusion in a Bird’s Eye View Plane. Sensors 2021.
31 Hadamard-FCN 96.26 % 93.32 % 95.63 % 96.90 % 4.86 % 3.10 % 0.02 s GPU @ 1.5 Ghz (Python)
M. Oeljeklaus: An Integrated Approach for Traffic Scene Understanding from Monocular Cameras. 2021.
32 StixelNet II 96.22 % 91.24 % 95.13 % 97.33 % 5.48 % 2.67 % 1.2 s 1 core @ 3.0 Ghz (Matlab + C/C++)
N. Garnett, S. Silberstein, S. Oron, E. Fetaya, U. Verner, A. Ayash, V. Goldner, R. Cohen, K. Horn and D. Levi: Real-time category-based and general obstacle detection for autonomous driving. 5th Workshop on Computer Vision for Road Scene Understanding and Autonomous Driving (CVRSUAD'17, IEEE-ICCV 2017 Workshop) 2017.
33 MultiNet code 96.15 % 95.36 % 95.79 % 96.51 % 4.67 % 3.49 % 0.17 s GPU @ 2.5 Ghz (Python + C/C++)
M. Teichmann, M. Weber, J. Zoellner, R. Cipolla and R. Urtasun: MultiNet: Real-time Joint Semantic Reasoning for Autonomous Driving. CoRR 2016.
34 HA-DeepLabv3+ 96.10 % 95.03 % 95.48 % 96.73 % 5.03 % 3.27 % 0.06 s 1 core @ 2.5 Ghz (Python)
R. Fan, H. Wang, P. Cai, J. Wu, M. Bocus, L. Qiao and M. Liu: Learning collision-free space detection from stereo images: Homography matrix brings better data augmentation. IEEE/ASME Transactions on Mechatronics 2021.
35 RBNet 96.06 % 93.49 % 95.80 % 96.31 % 4.64 % 3.69 % 0.18 s GPU @ 2.5 Ghz (Matlab + C/C++)
Z. Chen and Z. Chen: RBNet: A Deep Neural Network for Unified Road and Road Boundary Detection. International Conference on Neural Information Processing 2017.
36 LoDNN
This method makes use of Velodyne laser scans.
96.05 % 95.03 % 95.79 % 96.31 % 4.66 % 3.69 % 18 ms GPU @ 2.5 Ghz (Torch)
L. Caltagirone, S. Scheidegger, L. Svensson and M. Wahde: Fast LIDAR-based Road Detection Using Fully Convolutional Neural Networks. IEEE Intelligent Vehicles Symposium 2017.
37 TEDNet
This method makes use of Velodyne laser scans.
code 95.94 % 95.31 % 96.21 % 95.67 % 4.14 % 4.33 % 0.09 s GPU @ 2.5 Ghz (Python)
M. Bayón-Gutiérrez, M. García- Ordás, H. Alaiz Moretón, J. Aveleira-Mata, S. Rubio-Martín and J. Benítez-Andrades: TEDNet: Twin Encoder Decoder Neural Network for 2D Camera and LiDAR Road Detection. Logic Journal of the IGPL 2024.
38 RoadNet3 95.88 % 95.46 % 96.37 % 95.40 % 3.95 % 4.60 % 300 ms GPU @ GTX950M (Python +Tensorflow)
Y. Lyu, L. Bai and X. Huang: Road Segmentation using CNN and Distributed LSTM. 2019 IEEE International Symposium on Circuits and Systems (ISCAS) 2019.
39 Up-Conv-Poly code 95.52 % 92.86 % 95.37 % 95.67 % 5.10 % 4.33 % 0.08 s GPU @ 2.5 Ghz (Python + C/C++)
G. Oliveira, W. Burgard and T. Brox: Efficient Deep Methods for Monocular Road Segmentation. IROS 2016.
40 DEEP-DIG 95.45 % 95.41 % 95.49 % 95.41 % 4.96 % 4.59 % 0.14 s GPU @ 3.5 Ghz (Python + C/C++)
J. Muñoz-Bulnes, C. Fernandez, I. Parra, D. Fernández-Llorca and M. Sotelo: Deep Fully Convolutional Networks with Random Data Augmentation for Enhanced Generalization in Road Detection. Workshop on Deep Learning for Autonomous Driving on IEEE 20th International Conference on Intelligent Transportation Systems 2017.
41 OFA Net code 95.43 % 89.10 % 92.78 % 98.24 % 8.41 % 1.76 % 0.04 s GPU @ 1.5 Ghz (Python)
S. Zhang, Z. Zhang, L. Sun and W. Qin: One For All: A Mutual Enhancement Method for Object Detection and Semantic Segmentation. Applied Sciences 2019.
42 CLRD
This method makes use of Velodyne laser scans.
95.41 % 94.83 % 95.23 % 95.59 % 5.26 % 4.41 % 0.05 s GPU @ 2.5 Ghz (Python)
M. Bayón-Gutiérrez, J. Benítez- Andrades, S. Rubio-Martín, J. Aveleira-Mata, H. Alaiz-Moretón and M. García-Ordás: Roadway Detection Using Convolutional Neural Network Through Camera and LiDAR Data. Hybrid Artificial Intelligent Systems 2022.
43 HID-LS
This method makes use of Velodyne laser scans.
94.89 % 91.46 % 95.37 % 94.42 % 5.04 % 5.58 % 0.25 s 1 cores @ 3.0 Ghz (C/C++)
S. Gu, Y. Zhang, J. Yang and H. Kong: Lidar-based urban road detection by histograms of normalized inverse depths and line scanning. ECMR 2017.
S. Gu, Y. Zhang, X. Yuan, J. Yang, T. Wu and H. Kong: Histograms of the Normalized Inverse Depth and Line Scanning for Urban Road Detection. IEEE Trans. Intelligent Transportation Systems 2019.
44 ChipNet
This method makes use of Velodyne laser scans.
94.87 % 91.31 % 95.21 % 94.53 % 5.23 % 5.47 % 12 ms GPU @ 1.5 Ghz (Keras)
Y. Lyu, L. Bai and X. Huang: ChipNet: Real-Time LiDAR Processing for Drivable Region Segmentation on an FPGA. IEEE Transactions on Circuits and Systems I: Regular Papers 2019.
45 DDN 94.17 % 92.70 % 96.73 % 91.74 % 3.41 % 8.26 % 2 s GPU @ 2.5 Ghz (Python + C/C++)
R. Mohan: Deep Deconvolutional Networks for Scene Parsing. 2014.
46 FCN-LC 94.09 % 90.26 % 94.05 % 94.13 % 6.55 % 5.87 % 0.03 s GPU Titan X
C. Mendes, V. Frémont and D. Wolf: Exploiting Fully Convolutional Neural Networks for Fast Road Detection. IEEE Conference on Robotics and Automation (ICRA) 2016.
47 ALO-AVG-MM code 94.05 % 90.96 % 94.82 % 93.29 % 5.60 % 6.71 % 0.0296 sec GeForce GTX 1080 GPU (Python)
F. Reis, R. Almeida, E. Kijak, S. Malinowski, S. Guimaraes and Z. Jr.: Combining convolutional side-outputs for road image segmentation. 2019 International Joint Conference on Neural Networks (IJCNN) - \textbfAccepted 2019.
48 RoadNet-RT 93.98 % 95.19 % 94.47 % 93.49 % 6.01 % 6.51 % 8m s GPU @ 2.5 Ghz (Python)
L. Bai, Y. Lyu and X. Huang: RoadNet-RT: High Throughput CNN Architecture and SoC Design for Real-Time Road Segmentation. arXiv preprint arXiv:2006.07644 2020.
49 RGB36-Super 93.90 % 94.39 % 94.70 % 93.11 % 5.73 % 6.89 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
L. Caltagirone, S. Lennart, M. Wahde and M. Sanfridson: Lidar-Camera Co-Training for Semi- Supervised Road Detection. arXiv preprint arXiv:1911.12597 2019.
50 Up-Conv 93.89 % 92.62 % 94.57 % 93.22 % 5.89 % 6.78 % 0.05 s GPU @ 2.5 Ghz (C/C++)
G. Oliveira, W. Burgard and T. Brox: Efficient Deep Methods for Monocular Road Segmentation. IROS 2016.
51 HIM 93.55 % 90.38 % 94.18 % 92.92 % 6.31 % 7.08 % 7 s >8 cores @ 2.5 Ghz (Python + C/C++)
D. Munoz, J. Bagnell and M. Hebert: Stacked Hierarchical Labeling. European Conference on Computer Vision (ECCV) 2010.
52 LidarHisto
This method makes use of Velodyne laser scans.
code 93.32 % 93.19 % 95.39 % 91.34 % 4.85 % 8.66 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
L. Chen, J. Yang and H. Kong: Lidar-histogram for fast road and obstacle detection. 2017 IEEE International Conference on Robotics and Automation (ICRA) 2017.
53 StixelNet 93.26 % 87.15 % 90.63 % 96.06 % 10.92 % 3.94 % 1 s GPU @ 2.5 Ghz (C/C++)
D. Levi, N. Garnett and E. Fetaya: StixelNet: A Deep Convolutional Network for Obstacle Detection and Road Segmentation.. 26TH British Machine Vision Conference (BMVC) 2015.
54 FTP 92.98 % 92.89 % 91.84 % 94.15 % 9.20 % 5.85 % 0.28 s GPU @ 2.5 Ghz (C/C++)
A. Laddha, M. Kocamaz, L. Navarro-Serment and M. Hebert: Map-Supervised Road Detection. IEEE Intelligent Vehicles Symposium Proceedings 2016.
55 MixedCRF
This method makes use of Velodyne laser scans.
92.75 % 90.24 % 94.03 % 91.50 % 6.39 % 8.50 % 6s 1 core @ 2.5 Ghz (Matlab + C/C++)
X. Han, H. Wang, J. Lu and C. Zhao: Road detection based on the fusion of Lidar and image data. 2017.
56 BMCF
This method uses stereo information.
92.21 % 87.99 % 91.55 % 92.89 % 9.43 % 7.11 % 2.5 s 1 core @ 2.5 Ghz (C/C++)
L. Wang, T. Wu, Z. Xiao, L. Xiao, D. Zhao and J. Han: Multi-cue road boundary detection using stereo vision. 2016 IEEE International Conference on Vehicular Electronics and Safety (ICVES) 2016.
57 HybridCRF
This method makes use of Velodyne laser scans.
91.95 % 86.44 % 94.01 % 89.98 % 6.30 % 10.02 % 1.5 s 1 core @ 2.5 Ghz (C/C++)
L. Xiao, R. Wang, B. Dai, Y. Fang, D. Liu and T. Wu: Hybrid conditional random field based camera-LIDAR fusion for road detection. Information Sciences 2018.
58 PGM-ARS 91.76 % 84.80 % 88.05 % 95.80 % 14.30 % 4.20 % 0.05 s i74700MQ @ 2.1Ghz (C/C++)
M. Passani, J. Yebes and L. Bergasa: Fast Pixelwise Road Inference Based on Uniformly Reweighted Belief Propagation . Proc. IEEE Intelligent Vehicles Symposium 2015.
59 ProbBoost
This method uses stereo information.
91.36 % 84.92 % 88.18 % 94.78 % 13.97 % 5.22 % 2.5 min >8 cores @ 3.0 Ghz (C/C++)
G. Vitor, A. Victorino and J. Ferreira: A probabilistic distribution approach for the classification of urban roads in complex environments. Workshop on Modelling, Estimation, Perception and Control of All Terrain Mobile Robots on IEEE International Conference on Robotics and Automation (ICRA) 2014.
60 NNP
This method uses stereo information.
91.34 % 88.65 % 91.07 % 91.60 % 9.87 % 8.40 % 5 s 4 cores @ 2.5 Ghz (Matlab)
X. Chen, K. Kundu, Y. Zhu, A. Berneshawi, H. Ma, S. Fidler and R. Urtasun: 3D Object Proposals for Accurate Object Class Detection. NIPS 2015.
61 multi-task CNN 91.15 % 87.45 % 85.08 % 98.15 % 18.92 % 1.85 % 25.1 ms GPU @ 2.0 Ghz (Python)
M. Oeljeklaus, F. Hoffmann and T. Bertram: A Fast Multi-Task CNN for Spatial Understanding of Traffic Scenes. IEEE Intelligent Transportation Systems Conference 2018.
62 SRF 90.77 % 92.44 % 89.35 % 92.23 % 12.08 % 7.77 % 0.2 s 1 core @ 2.5 Ghz (C/C++)
L. Xiao, B. Dai, D. Liu, D. Zhao and T. Wu: Monocular Road Detection Using Structured Random Forest. Int J Adv Robot Syst 2016.
63 RES3D-Velo
This method makes use of Velodyne laser scans.
90.60 % 85.38 % 85.96 % 95.78 % 17.20 % 4.22 % 0.36 s 1 core @ 2.5 Ghz (C/C++)
P. Shinzato, D. Wolf and C. Stiller: Road Terrain Detection: Avoiding Common Obstacle Detection Assumptions Using Sensor Fusion. Intelligent Vehicles Symposium (IV) 2014.
64 CB 90.55 % 85.40 % 92.75 % 88.45 % 7.60 % 11.55 % 2 s 1 core @ 3.4 Ghz (Python) + GPU
C. Mendes, V. Frémont and D. Wolf: Vision-Based Road Detection using Contextual Blocks. 2015.
65 MAP 89.97 % 92.14 % 87.47 % 92.62 % 14.58 % 7.38 % 0.28s GPU
A. Laddha, M. Kocamaz, L. Navarro-Serment and M. Hebert: Map-Supervised Road Detection. IEEE Intelligent Vehicles Symposium Proceedings 2016.
66 SPRAY 89.69 % 93.84 % 89.13 % 90.25 % 12.10 % 9.75 % 45 ms NVIDIA GTX 580 (Python + OpenCL)
T. Kuehnl, F. Kummert and J. Fritsch: Spatial Ray Features for Real-Time Ego-Lane Extraction. Proc. IEEE Intelligent Transportation Systems 2012.
67 ARSL-AMI 89.56 % 82.82 % 85.87 % 93.59 % 16.93 % 6.41 % 0.05 s 4 cores @ 2.5 Ghz (C/C++)
M. Passani, J. Yebes and L. Bergasa: CRF-based semantic labeling in miniaturized road scenes . Proc. IEEE Intelligent Transportation Systems 2014.
68 FusedCRF
This method makes use of Velodyne laser scans.
89.51 % 83.53 % 86.64 % 92.58 % 15.69 % 7.42 % 2 s 1 core @ 2.5 Ghz (C/C++)
L. Xiao, B. Dai, D. Liu, T. Hu and T. Wu: CRF based Road Detection with Multi-Sensor Fusion. Intelligent Vehicles Symposium (IV) 2015.
69 BM
This method uses stereo information.
89.41 % 80.61 % 83.43 % 96.30 % 21.02 % 3.70 % 2 s 2 cores @ 2.5 Ghz (Matlab)
B. WANG, V. Fremont and S. Rodriguez Florez: Color-based Road Detection and its Evaluation on the KITTI Road Benchmark. Workshop on Benchmarking Road Terrain and Lane Detection Algorithms for In-Vehicle Application, IEEE Intelligent Vehicles Symposium 2014.
70 HistonBoost
This method uses stereo information.
88.73 % 81.57 % 84.49 % 93.42 % 18.85 % 6.58 % 2.5 min >8 cores @ 3.0 Ghz (C/C++)
G. Vitor, A. Victorino and J. Ferreira: Comprehensive Performance Analysis of Road Detection Algorithms Using the Common Urban Kitti-Road Benchmark. Workshop on Benchmarking Road Terrain and Lane Detection Algorithms for In-Vehicle Application on IEEE Intelligent Vehicles Symposium (IV) 2014.
71 geo+gpr+crf
This method uses stereo information.
88.20 % 82.33 % 85.32 % 91.27 % 17.26 % 8.73 % 30 s 1 core @ 2.0 Ghz (C/C++)
Z. Xiao, B. Dai, H. Li, T. Wu, X. Xu, Y. Zeng and T. Chen: Gaussian process regression-based robust free space detection for autonomous vehicle by 3-D point cloud and 2-D appearance information fusion. International Journal of Advanced Robotic Systems 2017.
72 GRES3D+VELO
This method makes use of Velodyne laser scans.
88.19 % 88.65 % 83.98 % 92.85 % 19.48 % 7.15 % 60 ms 4 cores @ 2.8 Ghz (C/C++)
P. Shinzato: Estimation of obstacles and road area with sparse 3D points. 2015.
73 SCRFFPFHGSP
This method uses stereo information.
87.96 % 83.16 % 90.01 % 86.01 % 10.50 % 13.99 % 5 s 8 cores @ 2.5 Ghz (C/C++, Matlab)
I. Gheorghe: Semantic Segmentation of Terrain and Road Terrain for Advanced Driver Assistance Systems. 2015.
74 GRES3D+SELAS
This method uses stereo information.
87.57 % 90.52 % 85.92 % 89.28 % 16.08 % 10.72 % 110 ms 4 core @ 2.8 Ghz (C/C++)
P. Shinzato: Estimation of obstacles and road area with sparse 3D points. 2015.
75 CN 86.21 % 84.40 % 82.85 % 89.86 % 20.45 % 10.14 % 2 s 1 core @ 2.5 Ghz (C/C++)
J. Alvarez, T. Gevers, Y. LeCun and A. Lopez: Road Scene Segmentation from a Single Image. ECCV 2012 2012.
76 RES3D-Stereo
This method uses stereo information.
83.62 % 85.74 % 79.81 % 87.81 % 24.42 % 12.19 % 0.7 s 1 core @ 2.5 Ghz (C/C++)
P. Shinzato, D. Gomes and D. Wolf: Road estimation with sparse 3D points from stereo data. Intelligent Transportation Systems (ITSC), 2014 IEEE 17th International Conference on 2014.
77 SPlane
This method uses stereo information.
82.28 % 82.83 % 76.85 % 88.53 % 29.32 % 11.47 % 2 s 1 core @ 3.0 Ghz (C/C++)
N. Einecke and J. Eggert: Block-Matching Stereo with Relaxed Fronto-Parallel Assumption. IV 2014.
78 SPlane + BL
This method uses stereo information.
82.04 % 85.56 % 75.11 % 90.39 % 32.93 % 9.61 % 2 s 1 core @ 3.0 Ghz (C/C++)
N. Einecke and J. Eggert: Block-Matching Stereo with Relaxed Fronto-Parallel Assumption. IV 2014.
79 ANN
This method uses stereo information.
80.95 % 68.36 % 69.95 % 96.05 % 45.35 % 3.95 % 3 s 1 core @ 3.0 Ghz (C/C++)
G. Vitor, D. Lima, A. Victorino and J. Ferreira: A 2D/3D Vision Based Approach Applied to Road Detection in Urban Environments. Intelligent Vehicles Symposium (IV), 2013 IEEE 2013.
Table as LaTeX | Only published Methods

UU_ROAD


Method Setting Code MaxF AP PRE REC FPR FNR Runtime Environment
1 SNE-RoadSegV2 97.08 % 92.87 % 96.83 % 97.34 % 1.04 % 2.66 % 0.03 s GPU @ 2.5 Ghz (Python)
Y. Feng, Y. Ma, Q. Chen, I. Pitas and R. Fan: SNE-RoadSegV2: Advancing Heterogeneous Feature Fusion and Fallibility Awareness for Freespace Detection. 2024.
2 UdeerLID+ 97.07 % 93.04 % 97.19 % 96.95 % 0.91 % 3.05 % 0.01 s 1 core @ 2.5 Ghz (C/C++)
T. Ni, X. Zhan, T. Luo, W. Liu, Z. Shi and J. Chen: UdeerLID+: Integrating LiDAR, Image, and Relative Depth with Semi-Supervised. 2024.
3 SNE-RoadSeg+ 97.04 % 92.97 % 96.84 % 97.24 % 1.03 % 2.76 % 0.08 s GPU @ 2.5 Ghz (Python)
H. Wang, R. Fan, P. Cai and M. Liu: SNE-RoadSeg+: Rethinking depth-normal translation and deep supervision for freespace detection. 2021 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) 2021.
4 RoadFormer+ 97.02 % 92.33 % 96.75 % 97.29 % 1.06 % 2.71 % 0.04 s 1 core @ 2.5 Ghz (C/C++)
J. Huang, J. Li, N. Jia, Y. Sun, C. Liu, Q. Chen and R. Fan: RoadFormer+: Delivering RGB-X Scene Parsing through Scale-Aware Information Decoupling and Advanced Heterogeneous Feature Fusion. IEEE Transactions on Intelligent Vehicles 2024.
5 RoadFormer 97.02 % 92.78 % 96.61 % 97.43 % 1.12 % 2.57 % 0.07 s GPU @ 2.5 Ghz (Python)
J. Li, Y. Zhang, P. Yun, G. Zhou, Q. Chen and R. Fan: RoadFormer: Duplex Transformer for RGB-Normal Semantic Road Scene Parsing. IEEE Transactions on Intelligent Vehicles 2024.
6 Pseudo-LiDAR 96.93 % 93.08 % 96.78 % 97.09 % 1.05 % 2.91 % 0.46 s GPU @ 2.5 Ghz (Python)
L. Sun, H. Zhang and W. Yin: Pseudo-LiDAR-Based Road Detection. IEEE Transactions on Circuits and Systems for Video Technology (T-CSVT) 2022.
7 LiDAR-DHMT 96.91 % 91.17 % 96.53 % 97.30 % 1.14 % 2.70 % 0.01 s 1 core @ 2.5 Ghz (Python)
8 DiPFormer 96.83 % 91.21 % 96.57 % 97.09 % 1.12 % 2.91 % 0.01 s 1 GPU @ 2.5 Ghz (C/C++)
S. Chen, T. Han, C. Zhang, W. Liu, J. Su, Z. Wang and G. Cai: Depth Matters: Exploring Deep Interactions of RGB-D for Semantic Segmentation in Traffic Scenes. arXiv preprint arXiv:2409.07995 2024.
9 Evi-RoadSeg code 96.52 % 92.43 % 95.94 % 97.10 % 1.34 % 2.90 % 0.01 s GPU @ 2.5 Ghz (Python)
F. Xue✝, Y. Chang✝, W. Xu, W. Liang, F. Sheng and A. Ming: Evidence-based Real-time Road Segmentation with RGB-D Data Augmentation. Transactions on Intelligent Transportation Systems (T-ITS) 2024.
10 UdeerLID 96.32 % 92.99 % 96.20 % 96.44 % 1.24 % 3.56 % 0.01 s 1 core @ 2.5 Ghz (C/C++)
T. Ni, X. Zhan, T. Luo, W. Liu, Z. Shi and J. Chen: UdeerLID+: Integrating LiDAR, Image, and Relative Depth with Semi-Supervised. 2024.
11 DFM-RTFNet 96.26 % 93.01 % 96.16 % 96.35 % 1.25 % 3.65 % 0.08 s GPU @ 2.5 Ghz (Python)
H. Wang, R. Fan, Y. Sun and M. Liu: Dynamic fusion module evolves drivable area and road anomaly detection: A benchmark and algorithms. IEEE Transactions on Cybernetics 2021.
12 LRDNet+
This method makes use of Velodyne laser scans.
code 96.18 % 90.03 % 95.94 % 96.42 % 1.33 % 3.58 % 0.01 s GPU @ 2.5 Ghz (Python)
A. Khan, J. Shao, Y. Rao, L. She and H. Shen: LRDNet: Lightweight LiDAR Aided Cascaded Feature Pools for Free Road Space Detection. IEEE Transactions on Multimedia 2025.
13 USNet code 96.11 % 91.71 % 95.86 % 96.37 % 1.36 % 3.63 % 0.02 s GPU @ 1.5 Ghz (Python)
Y. Chang, F. Xue, F. Sheng, W. Liang and A. Ming: Fast Road Segmentation via Uncertainty-aware Symmetric Network. IEEE International Conference on Robotics and Automation (ICRA) 2022.
14 LRDNet (L)
This method makes use of Velodyne laser scans.
code 96.10 % 89.59 % 95.97 % 96.22 % 1.32 % 3.78 % 0.1 s GPU @ 2.5 Ghz (Python)
A. Khan, J. Shao, Y. Rao, L. She and H. Shen: LRDNet: Lightweight LiDAR Aided Cascaded Feature Pools for Free Road Space Detection. IEEE Transactions on Multimedia 2025.
15 SNE-RoadSeg code 96.03 % 93.03 % 96.22 % 95.83 % 1.23 % 4.17 % 0.18 s GPU @ 2.5 Ghz (Python)
R. Fan, H. Wang, P. Cai and M. Liu: SNE-RoadSeg: Incorporating Surface Normal Information into Semantic Segmentation for Accurate Freespace Detection. ECCV 2020.
16 PLARD
This method makes use of Velodyne laser scans.
code 95.95 % 95.25 % 96.25 % 95.65 % 1.21 % 4.35 % 0.16 s GPU @ 2.5 Ghz (Python)
Z. Chen, J. Zhang and D. Tao: Progressive LiDAR adaptation for road detection. IEEE/CAA Journal of Automatica Sinica 2019.
17 LRDNet(S)
This method makes use of Velodyne laser scans.
code 95.78 % 90.75 % 95.62 % 95.95 % 1.43 % 4.05 % .009 s GPU @ 2.5 Ghz (Python)
A. Khan, J. Shao, Y. Rao, L. She and H. Shen: LRDNet: Lightweight LiDAR Aided Cascaded Feature Pools for Free Road Space Detection. IEEE Transactions on Multimedia 2025.
18 CLCFNet
This method makes use of Velodyne laser scans.
95.68 % 88.37 % 94.75 % 96.63 % 1.75 % 3.37 % 0.02 s GPU @ 1.5 Ghz (Python)
S. Gu, J. Yang and H. Kong: A Cascaded LiDAR-Camera Fusion Network for Road Detection. ICRA 2021.
19 3MT-RoadSeg
This method makes use of Velodyne laser scans.
This method makes use of GPS/IMU information.
code 95.63 % 92.77 % 95.37 % 95.89 % 1.52 % 4.11 % 0.07 s GPU @ 2.5 Ghz (Python)
E. Milli, �. Erkent and A. Yılmaz: Multi-Modal Multi-Task (3MT) Road Segmentation. IEEE Robotics and Automation Letters 2023.
20 MemRoadNet code 95.37 % 92.76 % 95.08 % 95.66 % 1.61 % 4.34 % 0.01 s GPU @ 2.5 Ghz (Python)
S. Shafiq, A. Khan and J. Shao: MemRoadNet: Human-like Memory Integration for Free Road Space Detection. Sensors 2025.
21 CLCFNet (LiDAR)
This method makes use of Velodyne laser scans.
95.25 % 88.02 % 94.36 % 96.16 % 1.87 % 3.84 % 0.02 s GPU @ 1.5 Ghz (C/C++)
S. Gu, J. Yang and H. Kong: A Cascaded LiDAR-Camera Fusion Network for Road Detection. ICRA 2021.
22 NIM-RTFNet 95.11 % 92.94 % 95.91 % 94.32 % 1.31 % 5.68 % 0.05 s GPU @ 2.5 Ghz (Python)
H. Wang, R. Fan, Y. Sun and M. Liu: Applying Surface Normal Information in Drivable Area and Road Anomaly Detection for Ground Mobile Robots. 2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) 2020.
23 RBANet 94.91 % 86.35 % 92.53 % 97.42 % 2.56 % 2.58 % 0.16 s GPU @ 1.5 Ghz (Python + C/C++)
J. Sun, S. Kim, S. Lee, Y. Kim and S. Ko: Reverse and Boundary Attention Network for Road Segmentation. Proceedings of the IEEE International Conference on Computer Vision Workshops 2019.
24 LidCamNet
This method makes use of Velodyne laser scans.
94.54 % 92.74 % 94.64 % 94.45 % 1.74 % 5.55 % 0.15 s GPU @ 2.5 Ghz (Python)
L. Caltagirone, M. Bellone, L. Svensson and M. Wahde: LIDAR-Camera Fusion for Road Detection Using Fully Convolutional Neural Networks. Robotics and Autonomous Systems 2018.
25 RGB36-Cotrain 94.53 % 92.54 % 94.60 % 94.46 % 1.76 % 5.54 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
L. Caltagirone, S. Lennart, M. Wahde and M. Sanfridson: Lidar-Camera Co-Training for Semi- Supervised Road Detection. arXiv preprint arXiv:1911.12597 2019.
26 SSLGAN 94.40 % 87.84 % 94.17 % 94.63 % 1.91 % 5.37 % 700ms GPU @ 1.5 Ghz (Python)
X. Han, J. Lu, C. Zhao, S. You and H. Li: Semisupervised and Weakly Supervised Road Detection Based on Generative Adversarial Networks. IEEE Signal Processing Letters 2018.
27 LC-CRF
This method makes use of Velodyne laser scans.
94.01 % 85.24 % 91.31 % 96.88 % 3.00 % 3.12 % 0.18 s GPU @ 1.5 Ghz (Python + C/C++)
S. Gu, Y. Zhang, J. Tang, J. Yang and H. Kong: Road Detection through CRF based LiDAR-Camera Fusion. ICRA 2019.
28 MultiNet code 93.69 % 92.55 % 94.24 % 93.14 % 1.85 % 6.86 % 0.17 s GPU @ 2.5 Ghz (Python + C/C++)
M. Teichmann, M. Weber, J. Zoellner, R. Cipolla and R. Urtasun: MultiNet: Real-time Joint Semantic Reasoning for Autonomous Driving. CoRR 2016.
29 TVFNet
This method makes use of Velodyne laser scans.
93.65 % 87.57 % 93.87 % 93.43 % 1.99 % 6.57 % 0.04 s GPU @ 1.5 Ghz (Python)
S. Gu, Y. Zhang, J. Yang, J. Alvarez and H. Kong: Two-View Fusion based Convolutional Neural Network for Urban Road Detection. IROS 2019.
30 LFD-RoadSeg code 93.49 % 92.19 % 93.46 % 93.52 % 2.13 % 6.48 % .004 s GPU @ 1.5 Ghz (Python)
H. Zhou, F. Xue, Y. Li, S. Gong, Y. Li and Y. Zhou: Exploiting Low-Level Representations for Ultra-Fast Road Segmentation. IEEE Transactions on Intelligent Transportation Systems 2024.
31 StixelNet II 93.40 % 85.01 % 91.05 % 95.87 % 3.07 % 4.13 % 1.2 s 1 core @ 3.0 Ghz (Matlab + C/C++)
N. Garnett, S. Silberstein, S. Oron, E. Fetaya, U. Verner, A. Ayash, V. Goldner, R. Cohen, K. Horn and D. Levi: Real-time category-based and general obstacle detection for autonomous driving. 5th Workshop on Computer Vision for Road Scene Understanding and Autonomous Driving (CVRSUAD'17, IEEE-ICCV 2017 Workshop) 2017.
32 HA-DeepLabv3+ 93.24 % 91.83 % 93.19 % 93.28 % 2.22 % 6.72 % 0.06 s 1 core @ 2.5 Ghz (Python)
R. Fan, H. Wang, P. Cai, J. Wu, M. Bocus, L. Qiao and M. Liu: Learning collision-free space detection from stereo images: Homography matrix brings better data augmentation. IEEE/ASME Transactions on Mechatronics 2021.
33 RBNet 93.21 % 89.18 % 92.81 % 93.60 % 2.36 % 6.40 % 0.18 s GPU @ 2.5 Ghz (Matlab + C/C++)
Z. Chen and Z. Chen: RBNet: A Deep Neural Network for Unified Road and Road Boundary Detection. International Conference on Neural Information Processing 2017.
34 Hadamard-FCN 93.14 % 90.00 % 93.31 % 92.98 % 2.17 % 7.02 % 0.02 s GPU @ 1.5 Ghz (Python)
M. Oeljeklaus: An Integrated Approach for Traffic Scene Understanding from Monocular Cameras. 2021.
35 BJN 93.13 % 87.59 % 93.89 % 92.38 % 1.96 % 7.62 % 0.02 s 1 core @ 2.5 Ghz (C/C++)
B. Yu, D. Lee, J. Lee and S. Kee: Free Space Detection Using Camera-LiDAR Fusion in a Bird’s Eye View Plane. Sensors 2021.
36 RoadNet3 92.95 % 91.93 % 93.32 % 92.58 % 2.16 % 7.42 % 300 ms GPU @ GTX950M (Python +Tensorflow)
Y. Lyu, L. Bai and X. Huang: Road Segmentation using CNN and Distributed LSTM. 2019 IEEE International Symposium on Circuits and Systems (ISCAS) 2019.
37 ChipNet
This method makes use of Velodyne laser scans.
92.91 % 84.95 % 90.98 % 94.91 % 3.06 % 5.09 % 12 ms GPU @ 1.5 Ghz (Keras)
Y. Lyu, L. Bai and X. Huang: ChipNet: Real-Time LiDAR Processing for Drivable Region Segmentation on an FPGA. IEEE Transactions on Circuits and Systems I: Regular Papers 2019.
38 TEDNet
This method makes use of Velodyne laser scans.
code 92.78 % 91.08 % 91.86 % 93.71 % 2.71 % 6.29 % 0.09 s GPU @ 2.5 Ghz (Python)
M. Bayón-Gutiérrez, M. García- Ordás, H. Alaiz Moretón, J. Aveleira-Mata, S. Rubio-Martín and J. Benítez-Andrades: TEDNet: Twin Encoder Decoder Neural Network for 2D Camera and LiDAR Road Detection. Logic Journal of the IGPL 2024.
39 Up-Conv-Poly code 92.65 % 89.20 % 92.85 % 92.45 % 2.32 % 7.55 % 0.08 s GPU @ 2.5 Ghz (Python + C/C++)
G. Oliveira, W. Burgard and T. Brox: Efficient Deep Methods for Monocular Road Segmentation. IROS 2016.
40 OFA Net code 92.62 % 83.12 % 88.97 % 96.58 % 3.90 % 3.42 % 0.04 s GPU @ 1.5 Ghz (Python)
S. Zhang, Z. Zhang, L. Sun and W. Qin: One For All: A Mutual Enhancement Method for Object Detection and Semantic Segmentation. Applied Sciences 2019.
41 CLRD
This method makes use of Velodyne laser scans.
92.41 % 90.16 % 92.30 % 92.52 % 2.52 % 7.48 % 0.05 s GPU @ 2.5 Ghz (Python)
M. Bayón-Gutiérrez, J. Benítez- Andrades, S. Rubio-Martín, J. Aveleira-Mata, H. Alaiz-Moretón and M. García-Ordás: Roadway Detection Using Convolutional Neural Network Through Camera and LiDAR Data. Hybrid Artificial Intelligent Systems 2022.
42 LoDNN
This method makes use of Velodyne laser scans.
92.29 % 90.35 % 90.81 % 93.81 % 3.09 % 6.19 % 18 ms GPU @ 2.5 Ghz (Torch)
L. Caltagirone, S. Scheidegger, L. Svensson and M. Wahde: Fast LIDAR-based Road Detection Using Fully Convolutional Neural Networks. IEEE Intelligent Vehicles Symposium 2017.
43 Up-Conv 91.89 % 89.44 % 92.59 % 91.20 % 2.38 % 8.80 % 0.05 s GPU @ 2.5 Ghz (C/C++)
G. Oliveira, W. Burgard and T. Brox: Efficient Deep Methods for Monocular Road Segmentation. IROS 2016.
44 DDN 91.76 % 86.84 % 93.06 % 90.50 % 2.20 % 9.50 % 2 s GPU @ 2.5 Ghz (Python + C/C++)
R. Mohan: Deep Deconvolutional Networks for Scene Parsing. 2014.
45 DEEP-DIG 91.27 % 91.77 % 91.32 % 91.22 % 2.82 % 8.78 % 0.14 s GPU @ 3.5 Ghz (Python + C/C++)
J. Muñoz-Bulnes, C. Fernandez, I. Parra, D. Fernández-Llorca and M. Sotelo: Deep Fully Convolutional Networks with Random Data Augmentation for Enhanced Generalization in Road Detection. Workshop on Deep Learning for Autonomous Driving on IEEE 20th International Conference on Intelligent Transportation Systems 2017.
46 RGB36-Super 91.15 % 90.16 % 89.68 % 92.68 % 3.48 % 7.32 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
L. Caltagirone, S. Lennart, M. Wahde and M. Sanfridson: Lidar-Camera Co-Training for Semi- Supervised Road Detection. arXiv preprint arXiv:1911.12597 2019.
47 RoadNet-RT 90.79 % 91.67 % 91.79 % 89.80 % 2.62 % 10.20 % 8m s GPU @ 2.5 Ghz (Python)
L. Bai, Y. Lyu and X. Huang: RoadNet-RT: High Throughput CNN Architecture and SoC Design for Real-Time Road Segmentation. arXiv preprint arXiv:2006.07644 2020.
48 HID-LS
This method makes use of Velodyne laser scans.
89.81 % 82.33 % 88.11 % 91.58 % 4.03 % 8.42 % 0.25 s 1 cores @ 3.0 Ghz (C/C++)
S. Gu, Y. Zhang, J. Yang and H. Kong: Lidar-based urban road detection by histograms of normalized inverse depths and line scanning. ECMR 2017.
S. Gu, Y. Zhang, X. Yuan, J. Yang, T. Wu and H. Kong: Histograms of the Normalized Inverse Depth and Line Scanning for Urban Road Detection. IEEE Trans. Intelligent Transportation Systems 2019.
49 FTP 89.62 % 88.93 % 89.10 % 90.14 % 3.59 % 9.86 % 0.28 s GPU @ 2.5 Ghz (C/C++)
A. Laddha, M. Kocamaz, L. Navarro-Serment and M. Hebert: Map-Supervised Road Detection. IEEE Intelligent Vehicles Symposium Proceedings 2016.
50 ALO-AVG-MM code 89.45 % 79.87 % 85.40 % 93.90 % 5.23 % 6.10 % 0.0296 sec GeForce GTX 1080 GPU (Python)
F. Reis, R. Almeida, E. Kijak, S. Malinowski, S. Guimaraes and Z. Jr.: Combining convolutional side-outputs for road image segmentation. 2019 International Joint Conference on Neural Networks (IJCNN) - \textbfAccepted 2019.
51 HybridCRF
This method makes use of Velodyne laser scans.
88.53 % 80.79 % 86.41 % 90.76 % 4.65 % 9.24 % 1.5 s 1 core @ 2.5 Ghz (C/C++)
L. Xiao, R. Wang, B. Dai, Y. Fang, D. Liu and T. Wu: Hybrid conditional random field based camera-LIDAR fusion for road detection. Information Sciences 2018.
52 LidarHisto
This method makes use of Velodyne laser scans.
code 86.55 % 81.13 % 90.71 % 82.75 % 2.76 % 17.25 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
L. Chen, J. Yang and H. Kong: Lidar-histogram for fast road and obstacle detection. 2017 IEEE International Conference on Robotics and Automation (ICRA) 2017.
53 FCN-LC 86.27 % 75.37 % 86.65 % 85.89 % 4.31 % 14.11 % 0.03 s GPU Titan X
C. Mendes, V. Frémont and D. Wolf: Exploiting Fully Convolutional Neural Networks for Fast Road Detection. IEEE Conference on Robotics and Automation (ICRA) 2016.
54 CB 86.13 % 75.21 % 86.47 % 85.80 % 4.38 % 14.20 % 2 s 1 core @ 3.4 Ghz (Python) + GPU
C. Mendes, V. Frémont and D. Wolf: Vision-Based Road Detection using Contextual Blocks. 2015.
55 StixelNet 86.06 % 72.05 % 82.61 % 89.82 % 6.16 % 10.18 % 1 s GPU @ 2.5 Ghz (C/C++)
D. Levi, N. Garnett and E. Fetaya: StixelNet: A Deep Convolutional Network for Obstacle Detection and Road Segmentation.. 26TH British Machine Vision Conference (BMVC) 2015.
56 HIM 85.76 % 76.18 % 87.65 % 83.95 % 3.86 % 16.05 % 7 s >8 cores @ 2.5 Ghz (Python + C/C++)
D. Munoz, J. Bagnell and M. Hebert: Stacked Hierarchical Labeling. European Conference on Computer Vision (ECCV) 2010.
57 MixedCRF
This method makes use of Velodyne laser scans.
85.69 % 75.12 % 80.17 % 92.02 % 7.42 % 7.98 % 6s 1 core @ 2.5 Ghz (Matlab + C/C++)
X. Han, H. Wang, J. Lu and C. Zhao: Road detection based on the fusion of Lidar and image data. 2017.
58 NNP
This method uses stereo information.
85.55 % 76.90 % 85.36 % 85.75 % 4.79 % 14.25 % 5 s 4 cores @ 2.5 Ghz (Matlab)
X. Chen, K. Kundu, Y. Zhu, A. Berneshawi, H. Ma, S. Fidler and R. Urtasun: 3D Object Proposals for Accurate Object Class Detection. NIPS 2015.
59 BMCF
This method uses stereo information.
85.46 % 74.07 % 85.06 % 85.86 % 4.91 % 14.14 % 2.5 s 1 core @ 2.5 Ghz (C/C++)
L. Wang, T. Wu, Z. Xiao, L. Xiao, D. Zhao and J. Han: Multi-cue road boundary detection using stereo vision. 2016 IEEE International Conference on Vehicular Electronics and Safety (ICVES) 2016.
60 FusedCRF
This method makes use of Velodyne laser scans.
84.49 % 72.35 % 77.13 % 93.40 % 9.02 % 6.60 % 2 s 1 core @ 2.5 Ghz (C/C++)
L. Xiao, B. Dai, D. Liu, T. Hu and T. Wu: CRF based Road Detection with Multi-Sensor Fusion. Intelligent Vehicles Symposium (IV) 2015.
61 MAP 84.44 % 87.17 % 83.66 % 85.23 % 5.42 % 14.77 % 0.28s GPU
A. Laddha, M. Kocamaz, L. Navarro-Serment and M. Hebert: Map-Supervised Road Detection. IEEE Intelligent Vehicles Symposium Proceedings 2016.
62 GRES3D+VELO
This method makes use of Velodyne laser scans.
84.14 % 80.20 % 80.57 % 88.03 % 6.92 % 11.97 % 60 ms 4 cores @ 2.8 Ghz (C/C++)
P. Shinzato: Estimation of obstacles and road area with sparse 3D points. 2015.
63 RES3D-Velo
This method makes use of Velodyne laser scans.
83.63 % 72.58 % 77.38 % 90.97 % 8.67 % 9.03 % 0.36 s 1 core @ 2.5 Ghz (C/C++)
P. Shinzato, D. Wolf and C. Stiller: Road Terrain Detection: Avoiding Common Obstacle Detection Assumptions Using Sensor Fusion. Intelligent Vehicles Symposium (IV) 2014.
64 SPRAY 82.71 % 87.19 % 82.16 % 83.26 % 5.89 % 16.74 % 45 ms NVIDIA GTX 580 (Python + OpenCL)
T. Kuehnl, F. Kummert and J. Fritsch: Spatial Ray Features for Real-Time Ego-Lane Extraction. Proc. IEEE Intelligent Transportation Systems 2012.
65 GRES3D+SELAS
This method uses stereo information.
82.70 % 83.95 % 78.54 % 87.32 % 7.77 % 12.68 % 110 ms 4 core @ 2.8 Ghz (C/C++)
P. Shinzato: Estimation of obstacles and road area with sparse 3D points. 2015.
66 geo+gpr+crf
This method uses stereo information.
81.00 % 69.74 % 79.78 % 82.27 % 6.79 % 17.73 % 30 s 1 core @ 2.0 Ghz (C/C++)
Z. Xiao, B. Dai, H. Li, T. Wu, X. Xu, Y. Zeng and T. Chen: Gaussian process regression-based robust free space detection for autonomous vehicle by 3-D point cloud and 2-D appearance information fusion. International Journal of Advanced Robotic Systems 2017.
67 SCRFFPFHGSP
This method uses stereo information.
80.78 % 70.80 % 81.07 % 80.50 % 6.13 % 19.50 % 5 s 8 cores @ 2.5 Ghz (C/C++, Matlab)
I. Gheorghe: Semantic Segmentation of Terrain and Road Terrain for Advanced Driver Assistance Systems. 2015.
68 ProbBoost
This method uses stereo information.
80.76 % 68.70 % 85.25 % 76.72 % 4.33 % 23.28 % 2.5 min >8 cores @ 3.0 Ghz (C/C++)
G. Vitor, A. Victorino and J. Ferreira: A probabilistic distribution approach for the classification of urban roads in complex environments. Workshop on Modelling, Estimation, Perception and Control of All Terrain Mobile Robots on IEEE International Conference on Robotics and Automation (ICRA) 2014.
69 multi-task CNN 80.45 % 75.87 % 68.63 % 97.19 % 14.48 % 2.81 % 25.1 ms GPU @ 2.0 Ghz (Python)
M. Oeljeklaus, F. Hoffmann and T. Bertram: A Fast Multi-Task CNN for Spatial Understanding of Traffic Scenes. IEEE Intelligent Transportation Systems Conference 2018.
70 PGM-ARS 79.94 % 67.77 % 77.37 % 82.67 % 7.88 % 17.33 % 0.05 s i74700MQ @ 2.1Ghz (C/C++)
M. Passani, J. Yebes and L. Bergasa: Fast Pixelwise Road Inference Based on Uniformly Reweighted Belief Propagation . Proc. IEEE Intelligent Vehicles Symposium 2015.
71 RES3D-Stereo
This method uses stereo information.
78.75 % 73.60 % 77.63 % 79.90 % 7.50 % 20.10 % 0.7 s 1 core @ 2.5 Ghz (C/C++)
P. Shinzato, D. Gomes and D. Wolf: Road estimation with sparse 3D points from stereo data. Intelligent Transportation Systems (ITSC), 2014 IEEE 17th International Conference on 2014.
72 BM
This method uses stereo information.
78.43 % 62.46 % 70.87 % 87.80 % 11.76 % 12.20 % 2 s 2 cores @ 2.5 Ghz (Matlab)
B. WANG, V. Fremont and S. Rodriguez Florez: Color-based Road Detection and its Evaluation on the KITTI Road Benchmark. Workshop on Benchmarking Road Terrain and Lane Detection Algorithms for In-Vehicle Application, IEEE Intelligent Vehicles Symposium 2014.
73 SRF 76.07 % 79.97 % 71.47 % 81.31 % 10.57 % 18.69 % 0.2 s 1 core @ 2.5 Ghz (C/C++)
L. Xiao, B. Dai, D. Liu, D. Zhao and T. Wu: Monocular Road Detection Using Structured Random Forest. Int J Adv Robot Syst 2016.
74 HistonBoost
This method uses stereo information.
74.19 % 63.01 % 77.43 % 71.22 % 6.77 % 28.78 % 2.5 min >8 cores @ 3.0 Ghz (C/C++)
G. Vitor, A. Victorino and J. Ferreira: Comprehensive Performance Analysis of Road Detection Algorithms Using the Common Urban Kitti-Road Benchmark. Workshop on Benchmarking Road Terrain and Lane Detection Algorithms for In-Vehicle Application on IEEE Intelligent Vehicles Symposium (IV) 2014.
75 SPlane + BL
This method uses stereo information.
74.02 % 79.61 % 65.15 % 85.68 % 14.93 % 14.32 % 2 s 1 core @ 3.0 Ghz (C/C++)
N. Einecke and J. Eggert: Block-Matching Stereo with Relaxed Fronto-Parallel Assumption. IV 2014.
76 SPlane
This method uses stereo information.
73.30 % 69.11 % 65.39 % 83.38 % 14.38 % 16.62 % 2 s 1 core @ 3.0 Ghz (C/C++)
N. Einecke and J. Eggert: Block-Matching Stereo with Relaxed Fronto-Parallel Assumption. IV 2014.
77 CN 72.25 % 66.61 % 71.96 % 72.54 % 9.21 % 27.46 % 2 s 1 core @ 2.5 Ghz (C/C++)
J. Alvarez, T. Gevers, Y. LeCun and A. Lopez: Road Scene Segmentation from a Single Image. ECCV 2012 2012.
78 ARSL-AMI 70.33 % 61.97 % 83.33 % 60.84 % 3.97 % 39.16 % 0.05 s 4 cores @ 2.5 Ghz (C/C++)
M. Passani, J. Yebes and L. Bergasa: CRF-based semantic labeling in miniaturized road scenes . Proc. IEEE Intelligent Transportation Systems 2014.
79 ANN
This method uses stereo information.
54.07 % 36.61 % 39.28 % 86.69 % 43.67 % 13.31 % 3 s 1 core @ 3.0 Ghz (C/C++)
G. Vitor, D. Lima, A. Victorino and J. Ferreira: A 2D/3D Vision Based Approach Applied to Road Detection in Urban Environments. Intelligent Vehicles Symposium (IV), 2013 IEEE 2013.
Table as LaTeX | Only published Methods

URBAN_ROAD


Method Setting Code MaxF AP PRE REC FPR FNR Runtime Environment
1 LiDAR-DHMT 97.59 % 92.93 % 97.33 % 97.84 % 1.48 % 2.16 % 0.01 s 1 core @ 2.5 Ghz (Python)
2 DiPFormer 97.57 % 92.94 % 97.34 % 97.79 % 1.47 % 2.21 % 0.01 s 1 GPU @ 2.5 Ghz (C/C++)
S. Chen, T. Han, C. Zhang, W. Liu, J. Su, Z. Wang and G. Cai: Depth Matters: Exploring Deep Interactions of RGB-D for Semantic Segmentation in Traffic Scenes. arXiv preprint arXiv:2409.07995 2024.
3 RoadFormer+ 97.56 % 93.74 % 97.43 % 97.69 % 1.42 % 2.31 % 0.04 s 1 core @ 2.5 Ghz (C/C++)
J. Huang, J. Li, N. Jia, Y. Sun, C. Liu, Q. Chen and R. Fan: RoadFormer+: Delivering RGB-X Scene Parsing through Scale-Aware Information Decoupling and Advanced Heterogeneous Feature Fusion. IEEE Transactions on Intelligent Vehicles 2024.
4 UdeerLID+ 97.55 % 93.98 % 97.46 % 97.65 % 1.40 % 2.35 % 0.01 s 1 core @ 2.5 Ghz (C/C++)
T. Ni, X. Zhan, T. Luo, W. Liu, Z. Shi and J. Chen: UdeerLID+: Integrating LiDAR, Image, and Relative Depth with Semi-Supervised. 2024.
5 SNE-RoadSegV2 97.55 % 93.98 % 97.57 % 97.53 % 1.34 % 2.47 % 0.03 s GPU @ 2.5 Ghz (Python)
Y. Feng, Y. Ma, Q. Chen, I. Pitas and R. Fan: SNE-RoadSegV2: Advancing Heterogeneous Feature Fusion and Fallibility Awareness for Freespace Detection. 2024.
6 RoadFormer 97.50 % 93.85 % 97.16 % 97.84 % 1.57 % 2.16 % 0.07 s GPU @ 2.5 Ghz (Python)
J. Li, Y. Zhang, P. Yun, G. Zhou, Q. Chen and R. Fan: RoadFormer: Duplex Transformer for RGB-Normal Semantic Road Scene Parsing. IEEE Transactions on Intelligent Vehicles 2024.
7 SNE-RoadSeg+ 97.50 % 93.98 % 97.41 % 97.58 % 1.43 % 2.42 % 0.08 s GPU @ 2.5 Ghz (Python)
H. Wang, R. Fan, P. Cai and M. Liu: SNE-RoadSeg+: Rethinking depth-normal translation and deep supervision for freespace detection. 2021 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) 2021.
8 Pseudo-LiDAR 97.42 % 94.09 % 97.30 % 97.54 % 1.49 % 2.46 % 0.46 s GPU @ 2.5 Ghz (Python)
L. Sun, H. Zhang and W. Yin: Pseudo-LiDAR-Based Road Detection. IEEE Transactions on Circuits and Systems for Video Technology (T-CSVT) 2022.
9 UdeerLID 97.22 % 94.02 % 97.15 % 97.29 % 1.57 % 2.71 % 0.01 s 1 core @ 2.5 Ghz (C/C++)
T. Ni, X. Zhan, T. Luo, W. Liu, Z. Shi and J. Chen: UdeerLID+: Integrating LiDAR, Image, and Relative Depth with Semi-Supervised. 2024.
10 Evi-RoadSeg code 97.08 % 93.54 % 96.57 % 97.59 % 1.91 % 2.41 % 0.01 s GPU @ 2.5 Ghz (Python)
F. Xue✝, Y. Chang✝, W. Xu, W. Liang, F. Sheng and A. Ming: Evidence-based Real-time Road Segmentation with RGB-D Data Augmentation. Transactions on Intelligent Transportation Systems (T-ITS) 2024.
11 PLARD
This method makes use of Velodyne laser scans.
code 97.03 % 94.03 % 97.19 % 96.88 % 1.54 % 3.12 % 0.16 s GPU @ 2.5 Ghz (Python)
Z. Chen, J. Zhang and D. Tao: Progressive LiDAR adaptation for road detection. IEEE/CAA Journal of Automatica Sinica 2019.
12 LRDNet+
This method makes use of Velodyne laser scans.
code 96.95 % 92.22 % 96.88 % 97.02 % 1.72 % 2.98 % 0.01 s GPU @ 2.5 Ghz (Python)
A. Khan, J. Shao, Y. Rao, L. She and H. Shen: LRDNet: Lightweight LiDAR Aided Cascaded Feature Pools for Free Road Space Detection. IEEE Transactions on Multimedia 2025.
13 USNet code 96.89 % 93.25 % 96.51 % 97.27 % 1.94 % 2.73 % 0.02 s GPU @ 1.5 Ghz (Python)
Y. Chang, F. Xue, F. Sheng, W. Liang and A. Ming: Fast Road Segmentation via Uncertainty-aware Symmetric Network. IEEE International Conference on Robotics and Automation (ICRA) 2022.
14 LRDNet (L)
This method makes use of Velodyne laser scans.
code 96.87 % 91.91 % 96.73 % 97.01 % 1.81 % 2.99 % 0.1 s GPU @ 2.5 Ghz (Python)
A. Khan, J. Shao, Y. Rao, L. She and H. Shen: LRDNet: Lightweight LiDAR Aided Cascaded Feature Pools for Free Road Space Detection. IEEE Transactions on Multimedia 2025.
15 DFM-RTFNet 96.78 % 94.05 % 96.62 % 96.93 % 1.87 % 3.07 % 0.08 s GPU @ 2.5 Ghz (Python)
H. Wang, R. Fan, Y. Sun and M. Liu: Dynamic fusion module evolves drivable area and road anomaly detection: A benchmark and algorithms. IEEE Transactions on Cybernetics 2021.
16 SNE-RoadSeg code 96.75 % 94.07 % 96.90 % 96.61 % 1.70 % 3.39 % 0.18 s GPU @ 2.5 Ghz (Python)
R. Fan, H. Wang, P. Cai and M. Liu: SNE-RoadSeg: Incorporating Surface Normal Information into Semantic Segmentation for Accurate Freespace Detection. ECCV 2020.
17 LRDNet(S)
This method makes use of Velodyne laser scans.
code 96.74 % 92.54 % 96.79 % 96.69 % 1.76 % 3.31 % .009 s GPU @ 2.5 Ghz (Python)
A. Khan, J. Shao, Y. Rao, L. She and H. Shen: LRDNet: Lightweight LiDAR Aided Cascaded Feature Pools for Free Road Space Detection. IEEE Transactions on Multimedia 2025.
18 MemRoadNet code 96.66 % 93.95 % 96.46 % 96.87 % 1.96 % 3.13 % 0.01 s GPU @ 2.5 Ghz (Python)
S. Shafiq, A. Khan and J. Shao: MemRoadNet: Human-like Memory Integration for Free Road Space Detection. Sensors 2025.
19 3MT-RoadSeg
This method makes use of Velodyne laser scans.
This method makes use of GPS/IMU information.
code 96.60 % 93.90 % 96.46 % 96.73 % 1.95 % 3.27 % 0.07 s GPU @ 2.5 Ghz (Python)
E. Milli, �. Erkent and A. Yılmaz: Multi-Modal Multi-Task (3MT) Road Segmentation. IEEE Robotics and Automation Letters 2023.
20 CLCFNet
This method makes use of Velodyne laser scans.
96.38 % 90.85 % 96.38 % 96.39 % 1.99 % 3.61 % 0.02 s GPU @ 1.5 Ghz (Python)
S. Gu, J. Yang and H. Kong: A Cascaded LiDAR-Camera Fusion Network for Road Detection. ICRA 2021.
21 RBANet 96.30 % 89.72 % 95.14 % 97.50 % 2.75 % 2.50 % 0.16 s GPU @ 1.5 Ghz (Python + C/C++)
J. Sun, S. Kim, S. Lee, Y. Kim and S. Ko: Reverse and Boundary Attention Network for Road Segmentation. Proceedings of the IEEE International Conference on Computer Vision Workshops 2019.
22 LidCamNet
This method makes use of Velodyne laser scans.
96.03 % 93.93 % 96.23 % 95.83 % 2.07 % 4.17 % 0.15 s GPU @ 2.5 Ghz (Python)
L. Caltagirone, M. Bellone, L. Svensson and M. Wahde: LIDAR-Camera Fusion for Road Detection Using Fully Convolutional Neural Networks. Robotics and Autonomous Systems 2018.
23 NIM-RTFNet 96.02 % 94.01 % 96.43 % 95.62 % 1.95 % 4.38 % 0.05 s GPU @ 2.5 Ghz (Python)
H. Wang, R. Fan, Y. Sun and M. Liu: Applying Surface Normal Information in Drivable Area and Road Anomaly Detection for Ground Mobile Robots. 2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) 2020.
24 CLCFNet (LiDAR)
This method makes use of Velodyne laser scans.
95.97 % 90.61 % 96.12 % 95.82 % 2.13 % 4.18 % 0.02 s GPU @ 1.5 Ghz (C/C++)
S. Gu, J. Yang and H. Kong: A Cascaded LiDAR-Camera Fusion Network for Road Detection. ICRA 2021.
25 LC-CRF
This method makes use of Velodyne laser scans.
95.68 % 88.34 % 93.62 % 97.83 % 3.67 % 2.17 % 0.18 s GPU @ 1.5 Ghz (Python + C/C++)
S. Gu, Y. Zhang, J. Tang, J. Yang and H. Kong: Road Detection through CRF based LiDAR-Camera Fusion. ICRA 2019.
26 RGB36-Cotrain 95.55 % 93.71 % 95.68 % 95.42 % 2.37 % 4.58 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
L. Caltagirone, S. Lennart, M. Wahde and M. Sanfridson: Lidar-Camera Co-Training for Semi- Supervised Road Detection. arXiv preprint arXiv:1911.12597 2019.
27 SSLGAN 95.53 % 90.35 % 95.84 % 95.24 % 2.28 % 4.76 % 700ms GPU @ 1.5 Ghz (Python)
X. Han, J. Lu, C. Zhao, S. You and H. Li: Semisupervised and Weakly Supervised Road Detection Based on Generative Adversarial Networks. IEEE Signal Processing Letters 2018.
28 TVFNet
This method makes use of Velodyne laser scans.
95.34 % 90.26 % 95.73 % 94.94 % 2.33 % 5.06 % 0.04 s GPU @ 1.5 Ghz (Python)
S. Gu, Y. Zhang, J. Yang, J. Alvarez and H. Kong: Two-View Fusion based Convolutional Neural Network for Urban Road Detection. IROS 2019.
29 LFD-RoadSeg code 95.21 % 93.71 % 95.35 % 95.08 % 2.56 % 4.92 % .004 s GPU @ 1.5 Ghz (Python)
H. Zhou, F. Xue, Y. Li, S. Gong, Y. Li and Y. Zhou: Exploiting Low-Level Representations for Ultra-Fast Road Segmentation. IEEE Transactions on Intelligent Transportation Systems 2024.
30 RBNet 94.97 % 91.49 % 94.94 % 95.01 % 2.79 % 4.99 % 0.18 s GPU @ 2.5 Ghz (Matlab + C/C++)
Z. Chen and Z. Chen: RBNet: A Deep Neural Network for Unified Road and Road Boundary Detection. International Conference on Neural Information Processing 2017.
31 BJN 94.89 % 90.63 % 96.14 % 93.67 % 2.07 % 6.33 % 0.02 s 1 core @ 2.5 Ghz (C/C++)
B. Yu, D. Lee, J. Lee and S. Kee: Free Space Detection Using Camera-LiDAR Fusion in a Bird’s Eye View Plane. Sensors 2021.
32 StixelNet II 94.88 % 87.75 % 92.97 % 96.87 % 4.04 % 3.13 % 1.2 s 1 core @ 3.0 Ghz (Matlab + C/C++)
N. Garnett, S. Silberstein, S. Oron, E. Fetaya, U. Verner, A. Ayash, V. Goldner, R. Cohen, K. Horn and D. Levi: Real-time category-based and general obstacle detection for autonomous driving. 5th Workshop on Computer Vision for Road Scene Understanding and Autonomous Driving (CVRSUAD'17, IEEE-ICCV 2017 Workshop) 2017.
33 MultiNet code 94.88 % 93.71 % 94.84 % 94.91 % 2.85 % 5.09 % 0.17 s GPU @ 2.5 Ghz (Python + C/C++)
M. Teichmann, M. Weber, J. Zoellner, R. Cipolla and R. Urtasun: MultiNet: Real-time Joint Semantic Reasoning for Autonomous Driving. CoRR 2016.
34 Hadamard-FCN 94.85 % 91.48 % 94.81 % 94.89 % 2.86 % 5.11 % 0.02 s GPU @ 1.5 Ghz (Python)
M. Oeljeklaus: An Integrated Approach for Traffic Scene Understanding from Monocular Cameras. 2021.
35 HA-DeepLabv3+ 94.83 % 93.24 % 94.77 % 94.89 % 2.88 % 5.11 % 0.06 s 1 core @ 2.5 Ghz (Python)
R. Fan, H. Wang, P. Cai, J. Wu, M. Bocus, L. Qiao and M. Liu: Learning collision-free space detection from stereo images: Homography matrix brings better data augmentation. IEEE/ASME Transactions on Mechatronics 2021.
36 TEDNet
This method makes use of Velodyne laser scans.
code 94.62 % 93.05 % 94.28 % 94.96 % 3.17 % 5.04 % 0.09 s GPU @ 2.5 Ghz (Python)
M. Bayón-Gutiérrez, M. García- Ordás, H. Alaiz Moretón, J. Aveleira-Mata, S. Rubio-Martín and J. Benítez-Andrades: TEDNet: Twin Encoder Decoder Neural Network for 2D Camera and LiDAR Road Detection. Logic Journal of the IGPL 2024.
37 RoadNet3 94.44 % 93.45 % 94.69 % 94.18 % 2.91 % 5.82 % 300 ms GPU @ GTX950M (Python +Tensorflow)
Y. Lyu, L. Bai and X. Huang: Road Segmentation using CNN and Distributed LSTM. 2019 IEEE International Symposium on Circuits and Systems (ISCAS) 2019.
38 CLRD
This method makes use of Velodyne laser scans.
94.20 % 92.66 % 94.25 % 94.14 % 3.16 % 5.86 % 0.05 s GPU @ 2.5 Ghz (Python)
M. Bayón-Gutiérrez, J. Benítez- Andrades, S. Rubio-Martín, J. Aveleira-Mata, H. Alaiz-Moretón and M. García-Ordás: Roadway Detection Using Convolutional Neural Network Through Camera and LiDAR Data. Hybrid Artificial Intelligent Systems 2022.
39 LoDNN
This method makes use of Velodyne laser scans.
94.07 % 92.03 % 92.81 % 95.37 % 4.07 % 4.63 % 18 ms GPU @ 2.5 Ghz (Torch)
L. Caltagirone, S. Scheidegger, L. Svensson and M. Wahde: Fast LIDAR-based Road Detection Using Fully Convolutional Neural Networks. IEEE Intelligent Vehicles Symposium 2017.
40 ChipNet
This method makes use of Velodyne laser scans.
94.05 % 88.29 % 93.57 % 94.53 % 3.58 % 5.47 % 12 ms GPU @ 1.5 Ghz (Keras)
Y. Lyu, L. Bai and X. Huang: ChipNet: Real-Time LiDAR Processing for Drivable Region Segmentation on an FPGA. IEEE Transactions on Circuits and Systems I: Regular Papers 2019.
41 DEEP-DIG 93.98 % 93.65 % 94.26 % 93.69 % 3.14 % 6.31 % 0.14 s GPU @ 3.5 Ghz (Python + C/C++)
J. Muñoz-Bulnes, C. Fernandez, I. Parra, D. Fernández-Llorca and M. Sotelo: Deep Fully Convolutional Networks with Random Data Augmentation for Enhanced Generalization in Road Detection. Workshop on Deep Learning for Autonomous Driving on IEEE 20th International Conference on Intelligent Transportation Systems 2017.
42 Up-Conv-Poly code 93.83 % 90.47 % 94.00 % 93.67 % 3.29 % 6.33 % 0.08 s GPU @ 2.5 Ghz (Python + C/C++)
G. Oliveira, W. Burgard and T. Brox: Efficient Deep Methods for Monocular Road Segmentation. IROS 2016.
43 OFA Net code 93.74 % 85.37 % 90.36 % 97.38 % 5.72 % 2.62 % 0.04 s GPU @ 1.5 Ghz (Python)
S. Zhang, Z. Zhang, L. Sun and W. Qin: One For All: A Mutual Enhancement Method for Object Detection and Semantic Segmentation. Applied Sciences 2019.
44 DDN 93.43 % 89.67 % 95.09 % 91.82 % 2.61 % 8.18 % 2 s GPU @ 2.5 Ghz (Python + C/C++)
R. Mohan: Deep Deconvolutional Networks for Scene Parsing. 2014.
45 HID-LS
This method makes use of Velodyne laser scans.
93.11 % 87.33 % 92.52 % 93.71 % 4.18 % 6.29 % 0.25 s 1 cores @ 3.0 Ghz (C/C++)
S. Gu, Y. Zhang, J. Yang and H. Kong: Lidar-based urban road detection by histograms of normalized inverse depths and line scanning. ECMR 2017.
S. Gu, Y. Zhang, X. Yuan, J. Yang, T. Wu and H. Kong: Histograms of the Normalized Inverse Depth and Line Scanning for Urban Road Detection. IEEE Trans. Intelligent Transportation Systems 2019.
46 RGB36-Super 92.94 % 92.29 % 93.14 % 92.74 % 3.77 % 7.26 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
L. Caltagirone, S. Lennart, M. Wahde and M. Sanfridson: Lidar-Camera Co-Training for Semi- Supervised Road Detection. arXiv preprint arXiv:1911.12597 2019.
47 RoadNet-RT 92.55 % 93.21 % 92.94 % 92.16 % 3.86 % 7.84 % 8m s GPU @ 2.5 Ghz (Python)
L. Bai, Y. Lyu and X. Huang: RoadNet-RT: High Throughput CNN Architecture and SoC Design for Real-Time Road Segmentation. arXiv preprint arXiv:2006.07644 2020.
48 Up-Conv 92.39 % 90.24 % 93.03 % 91.76 % 3.79 % 8.24 % 0.05 s GPU @ 2.5 Ghz (C/C++)
G. Oliveira, W. Burgard and T. Brox: Efficient Deep Methods for Monocular Road Segmentation. IROS 2016.
49 ALO-AVG-MM code 92.03 % 85.64 % 90.65 % 93.45 % 5.31 % 6.55 % 0.0296 sec GeForce GTX 1080 GPU (Python)
F. Reis, R. Almeida, E. Kijak, S. Malinowski, S. Guimaraes and Z. Jr.: Combining convolutional side-outputs for road image segmentation. 2019 International Joint Conference on Neural Networks (IJCNN) - \textbfAccepted 2019.
50 FTP 91.61 % 90.96 % 91.04 % 92.20 % 5.00 % 7.80 % 0.28 s GPU @ 2.5 Ghz (C/C++)
A. Laddha, M. Kocamaz, L. Navarro-Serment and M. Hebert: Map-Supervised Road Detection. IEEE Intelligent Vehicles Symposium Proceedings 2016.
51 HybridCRF
This method makes use of Velodyne laser scans.
90.81 % 86.01 % 91.05 % 90.57 % 4.90 % 9.43 % 1.5 s 1 core @ 2.5 Ghz (C/C++)
L. Xiao, R. Wang, B. Dai, Y. Fang, D. Liu and T. Wu: Hybrid conditional random field based camera-LIDAR fusion for road detection. Information Sciences 2018.
52 FCN-LC 90.79 % 85.83 % 90.87 % 90.72 % 5.02 % 9.28 % 0.03 s GPU Titan X
C. Mendes, V. Frémont and D. Wolf: Exploiting Fully Convolutional Neural Networks for Fast Road Detection. IEEE Conference on Robotics and Automation (ICRA) 2016.
53 LidarHisto
This method makes use of Velodyne laser scans.
code 90.67 % 84.79 % 93.06 % 88.41 % 3.63 % 11.59 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
L. Chen, J. Yang and H. Kong: Lidar-histogram for fast road and obstacle detection. 2017 IEEE International Conference on Robotics and Automation (ICRA) 2017.
54 HIM 90.64 % 81.42 % 91.62 % 89.68 % 4.52 % 10.32 % 7 s >8 cores @ 2.5 Ghz (Python + C/C++)
D. Munoz, J. Bagnell and M. Hebert: Stacked Hierarchical Labeling. European Conference on Computer Vision (ECCV) 2010.
55 MixedCRF
This method makes use of Velodyne laser scans.
90.59 % 84.24 % 89.11 % 92.13 % 6.20 % 7.87 % 6s 1 core @ 2.5 Ghz (Matlab + C/C++)
X. Han, H. Wang, J. Lu and C. Zhao: Road detection based on the fusion of Lidar and image data. 2017.
56 BMCF
This method uses stereo information.
89.75 % 84.15 % 89.02 % 90.49 % 6.15 % 9.51 % 2.5 s 1 core @ 2.5 Ghz (C/C++)
L. Wang, T. Wu, Z. Xiao, L. Xiao, D. Zhao and J. Han: Multi-cue road boundary detection using stereo vision. 2016 IEEE International Conference on Vehicular Electronics and Safety (ICVES) 2016.
57 NNP
This method uses stereo information.
89.68 % 86.50 % 89.67 % 89.68 % 5.69 % 10.32 % 5 s 4 cores @ 2.5 Ghz (Matlab)
X. Chen, K. Kundu, Y. Zhu, A. Berneshawi, H. Ma, S. Fidler and R. Urtasun: 3D Object Proposals for Accurate Object Class Detection. NIPS 2015.
58 StixelNet 89.12 % 81.23 % 85.80 % 92.71 % 8.45 % 7.29 % 1 s GPU @ 2.5 Ghz (C/C++)
D. Levi, N. Garnett and E. Fetaya: StixelNet: A Deep Convolutional Network for Obstacle Detection and Road Segmentation.. 26TH British Machine Vision Conference (BMVC) 2015.
59 CB 88.97 % 79.69 % 89.50 % 88.44 % 5.71 % 11.56 % 2 s 1 core @ 3.4 Ghz (Python) + GPU
C. Mendes, V. Frémont and D. Wolf: Vision-Based Road Detection using Contextual Blocks. 2015.
60 FusedCRF
This method makes use of Velodyne laser scans.
88.25 % 79.24 % 83.62 % 93.44 % 10.08 % 6.56 % 2 s 1 core @ 2.5 Ghz (C/C++)
L. Xiao, B. Dai, D. Liu, T. Hu and T. Wu: CRF based Road Detection with Multi-Sensor Fusion. Intelligent Vehicles Symposium (IV) 2015.
61 MAP 87.80 % 89.96 % 86.01 % 89.66 % 8.04 % 10.34 % 0.28s GPU
A. Laddha, M. Kocamaz, L. Navarro-Serment and M. Hebert: Map-Supervised Road Detection. IEEE Intelligent Vehicles Symposium Proceedings 2016.
62 ProbBoost
This method uses stereo information.
87.78 % 77.30 % 86.59 % 89.01 % 7.60 % 10.99 % 2.5 min >8 cores @ 3.0 Ghz (C/C++)
G. Vitor, A. Victorino and J. Ferreira: A probabilistic distribution approach for the classification of urban roads in complex environments. Workshop on Modelling, Estimation, Perception and Control of All Terrain Mobile Robots on IEEE International Conference on Robotics and Automation (ICRA) 2014.
63 SPRAY 87.09 % 91.12 % 87.10 % 87.08 % 7.10 % 12.92 % 45 ms NVIDIA GTX 580 (Python + OpenCL)
T. Kuehnl, F. Kummert and J. Fritsch: Spatial Ray Features for Real-Time Ego-Lane Extraction. Proc. IEEE Intelligent Transportation Systems 2012.
64 multi-task CNN 86.81 % 82.15 % 78.26 % 97.47 % 14.92 % 2.53 % 25.1 ms GPU @ 2.0 Ghz (Python)
M. Oeljeklaus, F. Hoffmann and T. Bertram: A Fast Multi-Task CNN for Spatial Understanding of Traffic Scenes. IEEE Intelligent Transportation Systems Conference 2018.
65 RES3D-Velo
This method makes use of Velodyne laser scans.
86.58 % 78.34 % 82.63 % 90.92 % 10.53 % 9.08 % 0.36 s 1 core @ 2.5 Ghz (C/C++)
P. Shinzato, D. Wolf and C. Stiller: Road Terrain Detection: Avoiding Common Obstacle Detection Assumptions Using Sensor Fusion. Intelligent Vehicles Symposium (IV) 2014.
66 GRES3D+VELO
This method makes use of Velodyne laser scans.
86.07 % 84.34 % 82.16 % 90.38 % 10.81 % 9.62 % 60 ms 4 cores @ 2.8 Ghz (C/C++)
P. Shinzato: Estimation of obstacles and road area with sparse 3D points. 2015.
67 PGM-ARS 85.69 % 73.83 % 82.34 % 89.33 % 10.56 % 10.67 % 0.05 s i74700MQ @ 2.1Ghz (C/C++)
M. Passani, J. Yebes and L. Bergasa: Fast Pixelwise Road Inference Based on Uniformly Reweighted Belief Propagation . Proc. IEEE Intelligent Vehicles Symposium 2015.
68 geo+gpr+crf
This method uses stereo information.
85.56 % 74.21 % 82.81 % 88.50 % 10.12 % 11.50 % 30 s 1 core @ 2.0 Ghz (C/C++)
Z. Xiao, B. Dai, H. Li, T. Wu, X. Xu, Y. Zeng and T. Chen: Gaussian process regression-based robust free space detection for autonomous vehicle by 3-D point cloud and 2-D appearance information fusion. International Journal of Advanced Robotic Systems 2017.
69 GRES3D+SELAS
This method uses stereo information.
85.09 % 86.86 % 82.27 % 88.10 % 10.46 % 11.90 % 110 ms 4 core @ 2.8 Ghz (C/C++)
P. Shinzato: Estimation of obstacles and road area with sparse 3D points. 2015.
70 SCRFFPFHGSP
This method uses stereo information.
84.93 % 76.31 % 85.37 % 84.49 % 7.98 % 15.51 % 5 s 8 cores @ 2.5 Ghz (C/C++, Matlab)
I. Gheorghe: Semantic Segmentation of Terrain and Road Terrain for Advanced Driver Assistance Systems. 2015.
71 HistonBoost
This method uses stereo information.
83.92 % 73.75 % 82.24 % 85.66 % 10.19 % 14.34 % 2.5 min >8 cores @ 3.0 Ghz (C/C++)
G. Vitor, A. Victorino and J. Ferreira: Comprehensive Performance Analysis of Road Detection Algorithms Using the Common Urban Kitti-Road Benchmark. Workshop on Benchmarking Road Terrain and Lane Detection Algorithms for In-Vehicle Application on IEEE Intelligent Vehicles Symposium (IV) 2014.
72 BM
This method uses stereo information.
83.47 % 72.23 % 75.90 % 92.72 % 16.22 % 7.28 % 2 s 2 cores @ 2.5 Ghz (Matlab)
B. WANG, V. Fremont and S. Rodriguez Florez: Color-based Road Detection and its Evaluation on the KITTI Road Benchmark. Workshop on Benchmarking Road Terrain and Lane Detection Algorithms for In-Vehicle Application, IEEE Intelligent Vehicles Symposium 2014.
73 SRF 82.44 % 87.37 % 80.60 % 84.36 % 11.18 % 15.64 % 0.2 s 1 core @ 2.5 Ghz (C/C++)
L. Xiao, B. Dai, D. Liu, D. Zhao and T. Wu: Monocular Road Detection Using Structured Random Forest. Int J Adv Robot Syst 2016.
74 RES3D-Stereo
This method uses stereo information.
81.08 % 81.68 % 78.14 % 84.24 % 12.98 % 15.76 % 0.7 s 1 core @ 2.5 Ghz (C/C++)
P. Shinzato, D. Gomes and D. Wolf: Road estimation with sparse 3D points from stereo data. Intelligent Transportation Systems (ITSC), 2014 IEEE 17th International Conference on 2014.
75 ARSL-AMI 80.36 % 70.23 % 83.24 % 77.67 % 8.61 % 22.33 % 0.05 s 4 cores @ 2.5 Ghz (C/C++)
M. Passani, J. Yebes and L. Bergasa: CRF-based semantic labeling in miniaturized road scenes . Proc. IEEE Intelligent Transportation Systems 2014.
76 SPlane + BL
This method uses stereo information.
79.63 % 83.90 % 72.59 % 88.17 % 18.34 % 11.83 % 2 s 1 core @ 3.0 Ghz (C/C++)
N. Einecke and J. Eggert: Block-Matching Stereo with Relaxed Fronto-Parallel Assumption. IV 2014.
77 CN 79.02 % 78.80 % 76.64 % 81.55 % 13.69 % 18.45 % 2 s 1 core @ 2.5 Ghz (C/C++)
J. Alvarez, T. Gevers, Y. LeCun and A. Lopez: Road Scene Segmentation from a Single Image. ECCV 2012 2012.
78 SPlane
This method uses stereo information.
78.69 % 77.16 % 71.96 % 86.80 % 18.63 % 13.20 % 2 s 1 core @ 3.0 Ghz (C/C++)
N. Einecke and J. Eggert: Block-Matching Stereo with Relaxed Fronto-Parallel Assumption. IV 2014.
79 ANN
This method uses stereo information.
67.70 % 52.50 % 54.19 % 90.17 % 41.98 % 9.83 % 3 s 1 core @ 3.0 Ghz (C/C++)
G. Vitor, D. Lima, A. Victorino and J. Ferreira: A 2D/3D Vision Based Approach Applied to Road Detection in Urban Environments. Intelligent Vehicles Symposium (IV), 2013 IEEE 2013.
Table as LaTeX | Only published Methods

Lane Estimation Evaluation

UM_LANE


Method Setting Code MaxF AP PRE REC FPR FNR Runtime Environment
1 CyberMELD+PLARD
This method makes use of Velodyne laser scans.
This method makes use of GPS/IMU information.
code 94.44 % 88.59 % 95.95 % 92.97 % 0.69 % 7.03 % 0.18 s 8 cores @ 1.5 Ghz (Python + C/C++)
X. Wang, Y. Qian, C. Wang and M. Yang: Map-Enhanced Ego-Lane Detection in the Missing Feature Scenarios. IEEE Access 2020.
2 CyberMELD
This method makes use of Velodyne laser scans.
This method makes use of GPS/IMU information.
code 93.56 % 88.58 % 95.94 % 91.30 % 0.68 % 8.70 % 0.05 s 8 core @ 1.5 Ghz (C/C++)
X. Wang, Y. Qian, C. Wang and M. Yang: Map-Enhanced Ego-Lane Detection in the Missing Feature Scenarios. IEEE Access 2020.
3 RoadNet3 91.47 % 91.01 % 91.78 % 91.17 % 1.44 % 8.83 % 300 ms GPU @ GTX950M (Python +Tensorflow)
Y. Lyu, L. Bai and X. Huang: Road Segmentation using CNN and Distributed LSTM. 2019 IEEE International Symposium on Circuits and Systems (ISCAS) 2019.
4 RBNet 90.54 % 82.03 % 94.92 % 86.56 % 0.82 % 13.44 % 0.18 s GPU @ 2.5 Ghz (Matlab + C/C++)
Z. Chen and Z. Chen: RBNet: A Deep Neural Network for Unified Road and Road Boundary Detection. International Conference on Neural Information Processing 2017.
5 Up-Conv-Poly code 89.88 % 87.52 % 92.01 % 87.84 % 1.34 % 12.16 % 0.08 s GPU @ 2.5 Ghz (Python + C/C++)
G. Oliveira, W. Burgard and T. Brox: Efficient Deep Methods for Monocular Road Segmentation. IROS 2016.
6 SPRAY 83.42 % 86.84 % 84.76 % 82.12 % 2.60 % 17.88 % 45 ms NVIDIA GTX 580 (Python + OpenCL)
T. Kuehnl, F. Kummert and J. Fritsch: Spatial Ray Features for Real-Time Ego-Lane Extraction. Proc. IEEE Intelligent Transportation Systems 2012.
7 SPlane + BL
This method uses stereo information.
69.63 % 73.78 % 80.01 % 61.63 % 2.71 % 38.37 % 2 s 1 core @ 3.0 Ghz (C/C++)
N. Einecke and J. Eggert: Block-Matching Stereo with Relaxed Fronto-Parallel Assumption. IV 2014.
8 SCRFFPFHGSP
This method uses stereo information.
57.22 % 39.34 % 41.78 % 90.79 % 22.28 % 9.21 % 5 s 8 cores @ 2.5 Ghz (C/C++, Matlab)
I. Gheorghe: Semantic Segmentation of Terrain and Road Terrain for Advanced Driver Assistance Systems. 2015.
Table as LaTeX | Only published Methods

Behaviour Evaluation

UM_LANE


Method Setting Code PRE-20 F1-20 HR-20 PRE-30 F1-30 HR-30 PRE-40 F1-40 HR-40 Runtime Environment
1 CyberMELD
This method makes use of Velodyne laser scans.
This method makes use of GPS/IMU information.
code 99.17 % 99.23 % 99.11 % 98.64 % 98.00 % 97.55 % 94.57 % 89.66 % 90.79 % 0.05 s 8 core @ 1.5 Ghz (C/C++)
X. Wang, Y. Qian, C. Wang and M. Yang: Map-Enhanced Ego-Lane Detection in the Missing Feature Scenarios. IEEE Access 2020.
2 CyberMELD+PLARD
This method makes use of Velodyne laser scans.
This method makes use of GPS/IMU information.
code 99.18 % 99.36 % 99.29 % 98.70 % 98.20 % 97.17 % 96.74 % 90.80 % 90.79 % 0.18 s 8 cores @ 1.5 Ghz (Python + C/C++)
X. Wang, Y. Qian, C. Wang and M. Yang: Map-Enhanced Ego-Lane Detection in the Missing Feature Scenarios. IEEE Access 2020.
3 RBNet 99.24 % 99.33 % 99.21 % 98.74 % 97.34 % 95.92 % 95.56 % 87.21 % 81.58 % 0.18 s GPU @ 2.5 Ghz (Matlab + C/C++)
Z. Chen and Z. Chen: RBNet: A Deep Neural Network for Unified Road and Road Boundary Detection. International Conference on Neural Information Processing 2017.
4 RoadNet3 99.18 % 99.21 % 99.07 % 98.39 % 97.23 % 95.57 % 94.57 % 83.72 % 80.26 % 300 ms GPU @ GTX950M (Python +Tensorflow)
Y. Lyu, L. Bai and X. Huang: Road Segmentation using CNN and Distributed LSTM. 2019 IEEE International Symposium on Circuits and Systems (ISCAS) 2019.
5 Up-Conv-Poly code 99.06 % 98.84 % 98.45 % 97.57 % 95.27 % 93.14 % 90.11 % 83.72 % 77.63 % 0.08 s GPU @ 2.5 Ghz (Python + C/C++)
G. Oliveira, W. Burgard and T. Brox: Efficient Deep Methods for Monocular Road Segmentation. IROS 2016.
6 SPRAY 97.58 % 96.74 % 96.38 % 96.59 % 94.16 % 92.06 % 87.64 % 78.57 % 62.16 % 45 ms NVIDIA GTX 580 (Python + OpenCL)
T. Kuehnl, F. Kummert and J. Fritsch: Spatial Ray Features for Real-Time Ego-Lane Extraction. Proc. IEEE Intelligent Transportation Systems 2012.
7 SPlane + BL
This method uses stereo information.
95.53 % 92.88 % 91.21 % 91.89 % 87.12 % 74.28 % 79.79 % 47.13 % 0.00 % 2 s 1 core @ 3.0 Ghz (C/C++)
N. Einecke and J. Eggert: Block-Matching Stereo with Relaxed Fronto-Parallel Assumption. IV 2014.
8 SCRFFPFHGSP
This method uses stereo information.
94.88 % 87.95 % 82.98 % 87.91 % 78.90 % 71.95 % 60.64 % 43.68 % 38.16 % 5 s 8 cores @ 2.5 Ghz (C/C++, Matlab)
I. Gheorghe: Semantic Segmentation of Terrain and Road Terrain for Advanced Driver Assistance Systems. 2015.


Related Datasets

Citation

When using this dataset in your research, we will be happy if you cite us:
@inproceedings{Fritsch2013ITSC,
author = {Jannik Fritsch and Tobias Kuehnl and Andreas Geiger},
title = {A New Performance Measure and Evaluation Benchmark for Road Detection Algorithms},
booktitle = {International Conference on Intelligent Transportation Systems (ITSC)},
year = {2013}
}

AltStyle によって変換されたページ (->オリジナル) /