Pedestrian Tracking based on Hybrid Structure Filters

Abstract

This paper proposed a object tracking method based on a hybrid filtering structure which includes Adaboost classifiers and particle filters to track a target automatically. The proposed algorithm first applied Adaboost classifier for object detection, filtering, and positioning of candidate targets, and then applied particle filter for confirming and tracking of the targets. With the help of Adaboost calculation for correction of the tracking results from particle filter, target missing events can be prevented efficiently. According to the experiment results, it is observed that compared with existing tracking methods, the performances of the proposed method in cases with disappeared objects, masked objects, and reappeared objects, in tracking target objects were much better.