A fast and accurate tracking approach for automated visual surveillance

Abstract

Currently, object detection and tracking as well as behavior analysis represent one of the main problems to be solved in automated visual surveillance. In this paper, a fast and accurate computer vision module that can track objects in unrestricted environments is described. The proposed approach is aimed at tracking arbitrary shapes on dynamic changing environments without any assumption on the nature and speed of the objects. The tracker approach exploits shape and motion information through a predicting-matching-updating paradigm. The described approach does not need a priori 2D model of the target object to be tracked.