Research On Vehicle Speed Detection Based On Monocular Video

As a basic and critical part of intelligent transportation system,vehicle speed detection provides the vehicle speed information for the system management in real time. Compared with traditional speed detection method, video speed detection has many advantages, such as low cost, easy installation and wide coverage. Especially, the monocular video speed detection is widely used because of its simple implementation,fast processing speed and low hardware cost. In this thesis, the method of monocular video speed detection is studied, and the camera calibration is carried out on this basis, the method of speed detection of virtual-loop in monocular video speed detection is also improved to improve its speed accuracy in some special cases. The specific work includes the following aspects:1. The method of monocular video speed detection is classified and summarized, and the main factors of error in the method of speed detection by tracking and by virtual-loop are analyzed.2. To calibrate camera using traffic lines. Firstly, detect the lane lines and the endpoints of lane boundaries in a single background image to mark the rectangular area of the road surface.Secondly, obtain the measurement information of the traffic signs and lines through the practical manual of "road traffic signs and markings". Finally, calibrate the camera accordingto the parallel and orthogonal relations contained in rectangular regions and the attributes of the vanishing points, so as to establish the mapping relationship between pixel coordinate and world coordinate to calculate the actual displacement of the vehicle.3. The method of virtual-loop speed detection is improved. The vehicle centroid tracking is added and multi loop detection mechanism is adopted to solve the two prominent problems of virtual-loop speed detection: (1) it is difficult to detect the vehicle when it changes lane; (2)when the vehicle speed is fast, it is easy to cause large speed error because of the fixed frame rate of video. In addition, the shadow removal method based on HSV color space is improved in the process of moving vehicle detection. Finally, the effectiveness of this method is verified by experiments.