Visual tracking with filtering algorithms

Abstract

We present a comparative study of object tracking algorithms using filtering methods. We detail the underlying model assumptions the different algorithms use, measure their operation performance, and compare them in real environmental settings. The comparison is based on several different criteria, including both the computational time and the performance of the tracker. We study a restricted family of methods, called filters. We compare the Kalman filter, unscented Kalman filter and the particle filtering methods. Based on real-world settings, some conclusions are drawn about the usability of the algorithms. We outline the conditions when a given algorithm becomes efficient.