A Vehicle Speed-measuring System Based On Binocular Stereo Vision

As the highway mileage increases dramatically, the vehicle over-speed phenomenaoccur frequently. In order to supervise the road safety, the traffic managing departmenthas paid much attention to vehicle-speed-measuring technology. The traditionalvideo-based vehicle speed measurement with a single camera has a few limits in spite ofbeing used. First, the prior knowledge of the environment is necessary. Second, thecamera must be fixed and cannot move a little after that.Binocular stereo vision theory has a wide range of applications in robot navigationand3D information acquisition. The object’s spatial position coordinates could bemeasured by a binocular stereo-vision system through two cameras without theinformation of the circumstances, so the binocular vision system has solved the twoproblems mentioned above. Therefore, we introduce stereo vision theory into the field.The main contents of this thesis are as follows:1. The existing video-based speed measurement technologies are investigated.With the foundation of monocular camera techniques and binocular stereo vision theory,a new method is put forward. Through theoretical analysis, we conclude that in carpositioning, our method could be accurate, real-time and practical.2. Binocular stereo vision models are studied, and factors influencing the precisionof the binocular measurement are expounded. At the same time, the relationshipbetween structural parameters and measurement errors is analyzed in detail. Throughthe comprehensive simulation of the structural parameters of binocular system, we cometo the basic conclusion: the bigger the lens’ focal length is, the greater the distance isbetween the two cameras, and the nearer the system is from the target, the betteraccuracy will be. Based on this conclusion, a binocular stereo vision system is designedand cameras’ calibration is finished.3. Image processing is a vital part in designing a new practical speed measurementsystem. Hence technologies of moving object detection and feature point extraction aresurveyed. After the analysis of the existing detection technologies and the feature pointmatching technologies, gaussian mixture model and SIFT feature extraction operator are adopted to detect vehicles and match feature points.4. Software program for the system is implemented based on MFC and OpenCV.Real vehicle motion videos are recorded to test the precision and reliability of themeasurement system. Abundant experiments with different vehicles under diverseenvironments have proved that the accuracy of binocular measurement system can meetthe expectation.