[Python] How to find the four corner coordinates of a black quadrilateral using OpenCV?

enter image description here

I want to find the corner coordinates of the two black quadrilaterals in the image.

It can be implemented in python or C++

My current method is based on “deep learning + OpenCV”, with the following pipeline:

NV12 frame input

（1）(NanoDet object detection model) → detection → confidence filtering → (x1, y1, x2, y2) … → sorting → [ROIs]

（2）(OpenCV) For each ROI: Gray enhancement→ Gaussian blur → Binarization → Morphological closing → findContours → take the largest contour → approxPolyDP (ε = 0.02 * perimeter) → keep only 4 points → cornerSubPix (sub-pixel refinement, suggested on gray ROI) → corner point sorting / rotation alignment → map back to full-image coordinates → save the 4 corner points of this ROI

Return: the set of quadrilateral corner points for all ROIs (square_results).

However, the actual performance is poor. How should I improve it?

Continue reading...