Abstract-A method of traffic flow detection for complex scenes based on image sequence is proposed, which is aiming at realizing intelligent vehicle detection and flow statistics for traffic videos shot by single camera located at the urban traffic intersection. For target vehicle detection, Gaussianfiltering mean method is utilized to create the dynamic realtime background; meanwhile it is combined with frame difference method to locate the motion vehicles in the foreground. Moreover, test-stripe detecting method is used for vehicle counting, in which data streams representing vehicle information are extracted with sliding window and are modified by Predictor-Corrector scheme to count the vehicles more accurately. Finally, a prototype system is designed to realize vehicle flow statistics for complex traffic scenes, combined with the judgment algorithm of vehicle traveling direction. Experimental simulation represents that target vehicle detection avoids the respective disadvantages of the two methods, that is background difference method and frame difference method, and obtains better detection effects. The prototype system captures good real-time performance and statistical data of the vehicle flow are comparatively accurate.