Revision history [back]

Well, your code is quite complicated, I didn't test it, and I don't understand all of it either. This is more a comment (but it's too long, so I have to put it as an answer). So a few ideas:

• keep your code simple, it's much easier to find a bug.
• use opencv routines to copy an image (frm.copyTo()) instead of std::move.
• capturing from a webcam doesn't take much time, so it isn't necessary to parallelize it*.

It will be much easier to capture the image, then do the processing in parallel, and when it's ready, capture the next frame. Here's a pseudocode:

Rect roi[10];
Mat frm(480,640,CV_8UC3);
Mat subfrm[10];
for(int i=0;i<10;i++)subfrm[i]=frm(roi[i]);  //define 10 ROIs on the image
while(1){
    cap>>frm;   //get the image
    for(int i=0;i<10;i++)thread[i].run(subfrm[i]);  //launch a thread for each ROI
    for(int i=0;i<10;i++)thread[i].join();  //wait for all the threads to finish
    imshow("Result",frm); //display the result
}

This code solves the problem of combining the frames, and saves some time because it won't make copies of the image. You won't need global variables or concurrency check (mutex).

*Of course, this isn't true anymore if you want to capture a large quantity of data (e.g. high resolution RAW image with high framerate) in a time-critical application.