Nonlinear Motion and Occlusion Handling for Pedestrian Tracking: Kalman Filter Approach

Abstract

This paper addresses the problem of multiple pedestrian tracking in realistic conditions for linear and non-linear motion of pedestrians with and without occlusion. The approach used involves a robust tracking algorithm using Kalman filter that connects tracking with detection to locate and maintain the identity of the pedestrians even during occlusions by taking the input measurement data from detection system for every frame. Kalman Filter is preferred for object tracking as it is flexible in case, motion of object is other than pure translational motion and performs better than other algorithms like mean-shift method in noisy atmosphere and noisy input. The Kalman filter used also acts as error corrector for limited number of erroneous input samples in measurement data. The tracking algorithm designed has been tested and verified for different cases of non-linear motion of pedestrians and occlusion between pedestrians for the custom data set and the comparison is reported.