Proof of concept
A reproducible Python/OpenCV script generates synthetic 720×440 component images, varies position and lighting, converts BGR→HSV, measures coloured-pixel ratios in five fixed ROIs (top_clip, lower_clip, seal, orientation, surface_mark) and returns PASS/FAIL — with REVIEW for extreme brightness (mean grey < 32 or > 156). It saves annotated images and a synthetic evaluation CSV/JSON.
No machine-learning model has been trained. Results on artificial images must not be described as accuracy on real factory components.
Output frames · seed 33
Rendered in your browser with the same engine as the simulator — like the OpenCV script's outputs/ folder.
In-browser evaluation
30 samples × 5 scenarios, brightness 0.75–1.25 (equivalent), jitter 6 px.
Reference run
Verified outputs from python qualiedge_poc.py --evaluate --count 30. Synthetic only.
Samples
150
| PASS | FAIL | REVIEW | |
|---|---|---|---|
| PASS | 30 | 0 | 0 |
| FAIL | 0 | 120 | 0 |
| Scenario | ROI ratios | Result |
|---|---|---|
| normal | top_clip 0.620 · lower_clip 0.588 · seal 0.424 · orientation 0.188 · surface_mark 0.000 | ✓ PASS |
| missing_clip | lower_clip 0.003 | ✕ FAIL |
| missing_seal | seal 0.065 | ✕ FAIL |
| wrong_orientation | orientation 0.000 | ✕ FAIL |
| surface_mark | surface_mark 0.155 | ✕ FAIL |
SYNTHETIC/ILLUSTRATIVE DATA ONLY. No machine learning model was trained.
Code excerpts
# HSV ranges (OpenCV: H 0–179, S/V 0–255) and fixed ROIs (x0, y0, x1, y1)
ROIS = {
class="s">'top_clip': (184, 139, 306, 177),
class="s">'lower_clip': (184, 249, 306, 288),
class="s">'seal': (468, 166, 545, 250),
class="s">'orientation': (140, 280, 183, 326),
class="s">'surface_mark': (315, 196, 409, 247),
}
HSV = {
class="s">'teal': ((80, 70, 60), (95, 255, 255)),
class="s">'orange': ((5, 80, 70), (25, 255, 255)),
class="s">'purple': ((120, 40, 55), (160, 255, 255)),
class="s">'red': [((0, 100, 100), (8, 255, 255)), ((170, 100, 100), (179, 255, 255))],
}def classify(image_bgr):
grey = cv2.cvtColor(image_bgr, cv2.COLOR_BGR2GRAY)
if grey.mean() < 32 or grey.mean() > 156:
return class="s">'REVIEW', [class="s">'Extreme brightness']
hsv = cv2.cvtColor(image_bgr, cv2.COLOR_BGR2HSV)
r = {k: ratio(hsv, ROIS[k], RULES[k].colour) for k in ROIS}
failed = [k for k, rule in RULES.items() if not rule.ok(r[k])]
return (class="s">'FAIL', failed) if failed else (class="s">'PASS', [])pip install -r requirements.txt # opencv-python, numpy
python qualiedge_poc.py # writes sample inputs + annotations to outputs/
python qualiedge_poc.py --evaluate --count 30