Features Selection for Multi-camera Tracking

Snatch theft is becoming more prevalent in Malaysia nowadays and proper measures must be taken to reduce it. CCTV surveillance systems have been widely used as a street crime prevention tool across public and private areas. Tracking the same object within different cameras´ view is essential in many surveillance applications. Recently, most of the researchers have grown more interest on how to track objects across cameras due to the increasing number of cameras. However, the current approach proposed by the researchers still offer trade-off in terms of its accuracy and speed. As the tracking accuracy increases, the speed will decrease that acts reversely proportional to it. This paper presents a novel approach to track moving objects across distributed cameras that provides the most optimal trade-off based on color, texture and edge features. The color, edge and texture features for target and candidate blobs are computed by a novel computational algorithm. This study focuses on analyzing of video surveillance in public places, specifically in outdoor environment. In the result section, the comparison between the effectiveness of the features used in the tracking algorithm is presented. The performance of the method is analyzed based on its accuracy and speed. The more suitable features are identified. Experimental results show the effectiveness of this method for real-time operation.