In order to reduce the possibility of accidents and injuries that may occur to a driving vehicle, developing vision assisted vehicles to improve the driving safety has become an active research topic among motor companies and computer vision communities around the world.

Tracking a moving vehicle with a moving camera is a challenging issue. The difficulties are caused by continuous changing of the camera position, variation of the appearance of the target in motion and  alteration of the illumination condition and background, etc. In particular, the camera-assisted car is required to react to the on-road situations in real time, which adds one more constraint to the tracker: the tracking algorithm has to be  computationally inexpensive.

The goal of this project is to develop a robust real-time tracking system, which pre-senses the vehicle crash so that the appropriate reaction to avoid it can be determined by some upper level applications. By continuously tracking a specific vehicle in front of the camera, one sudden growth in the size of the vehicle can be captured and a corresponding signal which infers a potential danger of car crash can be sent out to notify the driver.

The tracking method and results are presented in this web site. Currently there is only one vehicle in the front view can be tracked and the tracker acts independently from any detectors by performing a manual initialization step.