Thanks for the suggestion. However, currently my main concern is to extract the all possible pedestrian blob out of the real time video. And i would like to know some effective way to detect the blobs through real time video. This is due to some algorithms are not suit in the video and extra work need to be done in order to achieve the same result. Once all the blob is detected, the matching part will be done by the trained model. Probably using SVM + some boosting.
I just started on this project in a week time and i would like to implement it with Matlab in window and a normal USB camera first.