A real-time and robust approach for short-term multiple objects tracking

Abstract

A real-time and robust approach for short-term multiple objects tracking is proposed in this paper. In this method, motion detection is used to detect moving objects in fixed scenes. A special and efficient method of morphological operation is applied to filter noise and connect split objects by a window with user defined size. Object matching is done by nearest neighbor method based on distance associated with position, color histogram and gradient orientation histogram. A simple but efficient tracking method is also proposed. The experiment results demonstrate that our method is very robust to track objects and handle short-term occlusion. And, the computation cost of our approach is very low that high-level features can be added to our tracking method to enhance the tracking performance when long-term occlusion.