Real time multiple human tracking using Kalman Filter

Abstract

This paper presents a multilevel framework for multiple object tracking in simple and complex environments. The foreground object is obtained using Fuzzy Inference System (FIS) to deal with the illumination changes, shadows, repetitive motion of the objects and clutters in the scene. Multiple object tracking is performed using Hungarian Algorithm and Kalman Filter (KF). Kalman Filter provides an optimal estimate of its position at each time step. The optimality is guaranteed if all noise is Gaussian. KF gives better results based on position estimation to avoid occlusion. Hungarian Algorithm is used to find a particular human in successive frames. The multi-person tracking is a generalization of the single person tracker. We assume that the motion of each person is independent of others. For each object in the scene, a separate KF is initialized and models its trajectory.