Study on pedestrian detection and tracking with monocular vision

Abstract

Pedestrian detection is a rapidly evolving topic in many computer vision applications such as intelligent vehicle, surveillance and advanced robotics. The pedestrian collision avoidance not only requires detection of pedestrian but also requires prediction by tracking to analyze its dynamics and behaviors. The objective of this paper is to provide a method realizing pedestrian detection and tracking based on monocular vision. The first part of the paper is to detect pedestrian from the image. Both the rectangle features and edge orientation features are calculated by integral image techniques and Adaboost is used to fulfill discriminative features selection and classifiers training. The second part contains a pedestrian tracking method based on Kalman filtering. Experiments are performed to test and verify the pedestrian detection and tracking method under normal urban environments. The experiment results show that the method can detect and track pedestrian ahead of vehicle with different sizes and postures.