Closed circuit television (CCTV) systems deployed on roadway networks are believed to be among the best mechanisms for providing useful, reliable information for effective traffic management. Video Traffic Detection systems are now being considered key components of advanced traffic management systems. This study presents simple, yet effective algorithms for the detection of moving vehicles in real-time.
In detecting the existence of vehicles, the thesis develops three approaches : a gray-scale normalized correlation algorithm, a reflectance ratio change detection algorithm, and an edge magnitude ratio change detection algorithm. We experiment above 3 algorithms and search the best algorithm for detecting the existence of vehicles from real-world traffic scenes.
In the rear part of the work, we mainly take into account how to extract an initial shape boundary of a vehicle to be tracked. In fact, the extraction of the initial shape of a moving target is not only important but also difficult in the visual tracking areas.
In a series of extensive experiments in a real-world traffic scene, the proposed algorithm has achieved a 98% of success rate during daytime and even at night better than a 95% success rate.
