Vehicle geometric model is an important component of on-road vehicle detection system based on vision sensor. In order to improve robustness of on-road vehicle detection

the vehicle rear model in the world coordinate system is built using contour geometric feature of vehicle rear ahead; by coordinates transform

vehicle rear model in vision sensor coordinate system is built. And then the vehicle rear model in image is derived

where its parameter expressions and the correspondence relations between the model parameters in image and the rear geometric parameters of the vehicle ahead in the real world are given. This vehicle model formulates the relationships between vehicle rear image size and vehicle position in image. Utilizing it in the vehicle detection can suppress that the region no matching the on-road vehicle perspective is misrecognized as vehicle. Experiment shows the vehicle model can effectively improve robustness of vehicle detection.