| Accompanied by the rapid development of the national economy,Intelligent transportation systems has gradually become an important force to promote the development of modern transportation technology.Video traffic information detection technology with the advantages of easy to use,easy to maintain,large monitoring range,less impact on the environment,access to information and more quickly becomes a hot topic,and has greatly promoted the development of Intelligent transportation Systems.Video-based 3D vehicle tracking and vehicle speed detection technology is an important subject of intelligent transportation systems.We analyzed the research background and significance of vehicle tracking,3D vehicle tracking and vehicle speed detection.In view of the limitations of the video speed measurement algorithms,such as low speed measurement accuracy and high computational complexity,a 3D vehicle tracking and vehicle speed detection system is designed by using convolutional neural networks to track vehicle,The main work includes:Firstly,in terms of vehicle tracking,a vehicle tracking algorithm based on the Center Track multi-target tracking model is deeply studied.This method uses the Center Track network model to detect the center of the target in the current frame and predicts the displacement offset of the target center,and then uses a simple greedy matching algorithm to obtain the moving vehicle with the largest match between the upper and lower frames according to the output displacement offset,and marks the vehicle,complete the tracking of moving targets in continuous video frames.At the same time,the vehicle 3D information recognition algorithm based on camera parameter self-calibration in monocular traffic scenes is used to establish a monocular vision-based camera model and a stable road vanishing point calibration model according to typical traffic scenes in actual roads,which are used to calibrate cameras.parameters,and use the Center Track multi-target tracking model to centrally track the vehicles in the surveillance video.On this basis,a nonlinear optimization algorithm based on the vanishing point and diagonal constraints is proposed,and the 3D information recognition and 3D target tracking of the vehicle are completed by combining the calibration information..Secondly,a video speed detection method based on multiple intrusion lines is designed.This method lays out multiple intrusion lines with known relative distances in the video,then detects the number of frames when the vehicle passes through each intrusion line,and finally combines the number of frames,the sampling time of the camera,and the distance between the intrusion lines to generate the probability of the vehicle speed.Density function model to calculate vehicle speed.Through simulation and testing on the comprehensive data set Brno Comp Speed,the results show that the vehicle speed detection algorithm proposed in this thesis can effectively reduce the speed measurement error and has low computational complexity,and can be used as the core of the video vehicle speed detection system.Finally,3D vehicle tracking and vehicle speed detection system based on video stream is designed,the thesis conducts related research on the video speed measurement system,analyzed the difficulties existing in the system,and proposed related solutions to the difficulties,and designed a practical video speed measurement system,using real video and car simulation experiments to test system performance.The experimental results show that our designed system can effectively track the vehicle in the video.With the multi-intrusion line video speed detection algorithm,the calculated speed measurement error is controlled within a range of less than 3.2%,which meets the relevant speed measurement standards and has certain practical value. |
