Color-based box (and corner/parallelogram) detection utilities with multiple algorithm versions and Raspberry Pi deployment support.

• Multiple detection pipelines (V1.2 HSV + morphology, V3 channel subtraction).
• Configurable HSV color ranges via interactive ROI tools.
• Support for detecting 3 colored reference boxes + a yellow locator box.
• Parallelogram / corner arrangement utilities.
• Single-file deploy script for Raspberry Pi: raspi_functions.py.
• Modular refactored detector: color_detector_refactored.py.

• Core (versioned) scripts:
  • cbd_v1.2.py (HSV + distance transform)
  • cbd_v2.py
  • cbd_v3.py (newer experiments)
  • color_detector_refactored.py
  • raspi_functions.py (all-in-one)
• Tools (interactive config generation):
  • tools/create_config.py (main object colors)
  • tools/create_boxes_config.py (4 small boxes set)
  • tools/new_obj.py (extend existing config)
• Parallelogram helpers:
  • parallelogram_detector_dual_method.py
  • parallelogram_detector_v1_2.py

JSON configs are ignored by Git (.gitignore contains *.json) so generate them locally.

Main: config.json
Secondary (4-box mode): config_boxes.json

Both contain:

{
 "color_ranges": {
 "red": [ [ [low_H,low_S,low_V], [high_H,high_S,high_V] ], ... ],
 "green": [...],
 "blue": [...],
 "yellow": [...]
 },
 "big_box_crop": [x, y, w, h] (optional)
 "boxes_crop": [x, y, w, h] (in config_boxes.json)
}

Loaded via:

• load_config
• load_config (two variants: one for config.json near line ~227 and one for config_boxes.json near line ~377)
• load_config

If missing, some scripts fall back to hard‐coded defaults.

Interactive (requires GUI display):

python tools/create_config.py # builds config.json
python tools/create_boxes_config.py # builds config_boxes.json
python tools/new_obj.py # append/merge new object-specific ranges

Follow on-screen ROI selections. Color ranges are auto-generated per ROI.

Implemented in:

• preprocess_v1_2_method
• Supporting utilities: detect_contours_and_centroids

Flow:

1. HSV threshold per color range.
2. Morphological cleanup.
3. Distance transform to tighten mask.
4. Contour filtering (area + aspect ratio).
5. Representative centroid per channel.

Implemented in:

• preprocess_v3
• Inline helper: isolate_and_subtract_channel inside that function and standalone in raspi_functions.py (top section).

Flow:

1. Split BGR.
2. For target channel c: subtract other channels → enhances dominance.
3. Otsu threshold.
4. Morph open + close.
5. Contours extracted from cleaned mask.

Corner arrangement utilities (e.g., identifying relative positions) live around mid/lower sections of raspi_functions.py (see functions near detection of yellow + ordering, e.g., identify_corner_arrangement).

Use raspi_functions.py for minimal dependency deployment (aggregation of all logic). Copy only needed image + config assets beside the script.

1. Create a configuration file (config.json) in the project root. Use the interactive tool:

python tools/create_config.py

(Place the generated config.json next to the scripts.)

2. HSV-only detection (simple pipeline):

python cbd_v3.py

This uses HSV ranges from config.json.

3. Channel isolation + HSV (combined method):

python color_detector_refactored.py

Provide an image path; the script loads config.json automatically.

Notes:

• Regenerate config.json if lighting changes.
• Keep config.json in the same directory as the scripts.

• Start with evenly lit images.
• Avoid over-tight ranges; include slight illumination variance.
• Re-generate after lighting/environment changes.
• Yellow often needs separate (smaller) morphology kernel (MORPH_KERNEL_YELLOW in raspi_functions.py).

1. Add new color: update generation tool to prompt for it.
2. Add algorithm variant: create preprocess_vX function and mirror usage pattern.
3. Unify duplicate load_config functions if maintenance becomes difficult.

Install (typical):

pip install opencv-python numpy

(If running on Raspberry Pi, use opencv-contrib-python)

• Reduce resolution before processing for speed.
• Tune morphology kernel sizes (MORPH_KERNEL_COLOR, MORPH_KERNEL_YELLOW) per camera.
• Cache configs; avoid reloading each frame.

• Blank masks: verify HSV range order and lighting.
• Wrong color dominance in V3: ensure channel subtraction not saturating (inspect intermediate single-channel result).
• No contours: relax min area thresholds in functions like detect_contours_and_centroids.