Particle filter based multi-pedestrian tracking by HOG and HOF

Abstract

Automatic pedestrian detection and tracking is an important issue in the field of computer vision and robot navigation. We propose a scheme to implement multi-pedestrian tracking in a scene obtained by a static camera. We combine HOG and HOF features to describe the characteristics of persons. AdaBoost algorithm is then utilized to train a strong classifier for better detection accuracy of persons. We use particle filter as the tracking framework and train a online SVM classifier, which is the observation model, by reliable samples from associated detections without occlusion. In consideration of the target's velocity into the weights calculation, the data association is more reliable. The preliminary experiments on some benchmark data demonstrate the feasibility of the proposed scheme.