Research On Vehicle Detection And Tracking Based On SSD

With the development of urban society,the number of motor vehicles is increasing day by day,and the road congestion in cities is getting worse.The application of ITS has been paid more and more attention by everyone.The detection and tracking of vehicles is an important part of it.In practice,the vehicle detection algorithm needs to detect the vehicles appearing in the image or the video sequence frame in real time to obtain the target parameters,and transmit the detected vehicle information to the vehicle tracking algorithm,which tracks the vehicle to determine the position of the vehicle,Effectively track the vehicle in real time.Convention vehicle detection methods,such as optical flow method,inter-frame difference method,the test results unsatisfactory,it is difficult to meet the needs of the actual scene.At present,deep learning technology has become the cutting-edge technology in the direction of machine learning,and has achieved amazing results in many fields of research.In this dissertation,the traditional vehicle detection algorithm is analyzed.In view of its shortcomings,the vehicle detector based on multi-scale feature map is designed using depth learning technology.Then the vehicle detector based on multi-scale feature map Combined with Camshift tracking and Kalman filtering algorithms,a vehicle detection and tracking system is constructed.This article specific work is as follows:1.Aiming at the problems existing in the traditional vehicle detection algorithm,this article designs a multi-scale prediction SSD vehicle detector based on the features of different convolutional layers.The basic idea is to use forward-propagating CNN network to generate a series of fixed The size of the set of bounding boxes,and the score of the target category in the box,the final test is then obtained by the non-maximum suppression algorithm.In this article,the problem of vehicle detection is taken as a regression problem.During the model training process,the entire image is trained and then the test results are optimized.The experimental results show that the vehicle detector based on SSD has a good detection effect both in recognition performance and in time efficiency.2.The multi-scale feature map prediction SSD vehicle detector combined with Camshift tracking and Kalman filtering algorithm,designed and developed a vehicle detection and tracking system.This system mainly includes the motion detection module,the vehicle detection module and the vehicle tracking module.Vehicle detection and tracking system first video background before and after separation,the video in the motion region extracted,and then use the SSD vehicle detector based on the extraction of the vehicle movement detection area,and finally the use of Camshift tracking and Kalman filter algorithm to achieve the vehicle Real-time tracking.The test results show that the vehicle detection and tracking system designed in this article can meet the needs of the practical application of the function and performance.