In Intelligence Traffic System (ITS)

the accurate detection of traffic flux at real time on road scene is very important steps. The still camera based video detection approach is one of the most important methods with much superiority to others. In this paper

a knowledge based video traffic flux detection system is presented. The traffic segmentation and recognition is the main algorithm of this system. According to the knowledge of vehicles' movement

vehicles have huge inertia

we assure that vehicles are moving at straight line in short period. With this premise

we reproject the movement of features to the road plane. The reprojected velocity of a points of the vehicle have a proportion with the height to the road plane. In the detection system

the Canny edge feature was used to fit a straight line and to match in image sequences. In the recognition stage

we assign a vehicle model and the height to the type of vehicle and reconstruct the 3D space model and verify this model with rules. The experiment results show that this algorithm is better than previous method to detect vehicle in congest

occlusion and shadow cases.