Pedestrian Detection and Tracking Using HOG and Oriented-LBP Features

During the last decade, various successful human detection methods have been developed. However, most of these methods focus on finding powerful features or classifiers to obtain high detection rate. In this work we introduce a pedestrian detection and tracking system to extract and track human objectives using an on board monocular camera. The system is composed of three stages. A pedestrian detector, which is based on the non-overlap HOG feature and an Oriented LBP feature, is applied to find possible locations of humans. Then an object validation step verifies detection results and rejects false positives by using a temporal coherence condition. Finally, Kalman filtering is used to track detected pedestrians. For a 320x240 image, the implementation of the proposed system runs at about 14 frames/second, while maintaining an human detection rate similar to existing methods.