Detecting and Tracking Volunteers in Expo Videos

Abstract

A method is introduced for detecting and tracking volunteers in expo surveillance videos. A detection box slides on the images from the videos from left to right and then from up to down. Since the volunteers wear the green clothes, only the patches of enough green pixels are passed for further processing. Then, three kinds of features i.e. RGB color features, SURF features and HOG features are extracted from the patches. These features are sent to a trained cascade boosting classifier to determine whether the patches are the volunteers. In the experiments, the proposed method demonstrates good performances on three videos.