问题
I was looking into Kanade Lucas Tomasi Tracker in the following link. However, I was wondering how the KLT recognizes the new people have entered scene. I know that there is replenishing of bounding boxes every 10 frames, but in case a person say entered in the 5th frame. Does it mean the for his first 5th consecutive frames the person is not captured? If I understand that this threshold can be varied, however there will still be cases when this will happen whats so ever.
Can someone explain if I am getting this wrong please?
回答1:
Yes, you are correct. A new person will not be detected until the next time vision.CascadeObjectDetector is run. This example essentially assumes that when new people enter the field of view, they will stay visible for some time, and will be detected eventually.
It may be helpful to think about it in relation to the frame rate. If the code runs at 10 fps, then the re-detection occurs every second. In other words, it will take on the order of a second for a new face to be detected. If that is not acceptable, then you can try re-detecting more often. The thing to keep in mind, though, is that face detection is slower than KLT (vision.PointTracker), so if you run the face detector more often, your frame rate will decrease.
来源：https://stackoverflow.com/questions/29201984/kanade-lucas-tomasi-tracker