The trick is to use “cv2.CAP_ANY”

Python
1

I had been looking for a solution to a similar problem for a while but couldn’t find anything and almost gave up. Recently, I got back to the project and finally figured it out. Here is the solution: the trick is to use the right Capture API backend. I connected up to three cameras via one USB-C hub, and it worked. The trick is to use “cv2.CAP_ANY”.

Python opencv camera

import cv2 as cv
import numpy as np

def initialize_cameras():
    return [cv.VideoCapture(i, cv.CAP_ANY) for i in range(3)]

def release_cameras(cameras):
    for cap in cameras:
        cap.release()

def read_frames(cameras):
    return [cap.read() for cap in cameras]

def all_frames_valid(frames):
    return all(ret for ret, _ in frames)

def concatenate_frames(frames):
    v_divider = np.ones((frames[0][1].shape[0], 6, 3), dtype=np.uint8) * 255
    combined_frame = np.concatenate([frame for _, frame in frames], axis=1)
    return np.concatenate([combined_frame, v_divider], axis=1)

def main():
    cameras = initialize_cameras()
    run = True

    while run:
        frames = read_frames(cameras)
        
        if all_frames_valid(frames):
            combined_frame = concatenate_frames(frames)
            cv.imshow('TEST', combined_frame)
        else:
            release_cameras(cameras)
            cameras = initialize_cameras()
        
        if cv.waitKey(200) & 0xFF == ord('q'):
            release_cameras(cameras)
            run = False
            cv.destroyAllWindows()

if __name__ == "__main__":
    main()
3

using CAP_ANY is no trick at all.

that’s the default.

by the way, what is KLA?

4

You are right. I had used CAP_DSHOW and it didn’t work. Then I switched to CAP_ANY. If you don’t specify the backend the default would be CAP_ANY.