Vehicle extraction based on focus of attention, multi feature segmentation and tracking

In this paper, we propose an automatic object tracking method for traffic video surveillance. The method is designed as a hybrid between a region based and a feature based technique. In a first stage, a motion detection algorithm identifies the objects from the background providing binary masks of the moving objects. In a second stage, a segmentation tool based on a multi-feature analysis further segments the areas corresponding to moving objects into homogenous regions. For each region, the method provides a set of characteristic feature values, which are used to track the regions (and thus the objects) along time. The results of this low-level analysis can be exploited by the content understanding module of an advanced video surveillance system for the detection of potentially dangerous situations, for law enforcement purposes, and for statistical traffic analysis.