Bio Inspired Pedestrian Detection and Tracking

Abstract

Pedestrian detection and tracking remains a popular issue in computer vision, spawning many applications in robotics, surveillance and security, biometrics and humancomputer interaction. In this paper we present a biological framework for detecting and tracking pedestrians by using a monocular moving camera. This framework is based on visual cortex cells, namely complex and end-stopped cells, the last being used to extract keypoints. By employing a modified HOG descriptor combined with the responses of complex cells and a linear SVM, pedestrians can be detected. By combining the above information with keypoints, motion information and tracked features, persons can be tracked in complex scenarios where partial occlusions exist.