Improve The Application Of ViBe Algorithm And SSD In The Detection And Tracking Of Moving Vehicle Targets

With the development of urban society,the number of motor vehicles is increasing day by day,the urban road congestion is increasingly serious,the application of intelligent transportation system is increasingly widely concerned,vehicle detection and tracking is an important part of its.In practical applications,the vehicle detection algorithm needs to detect the vehicle appearing in the image or video sequence frame in real time,obtain the target parameters,and transmit the detected vehicle information to the vehicle tracking algorithm,which can track the vehicle to determine the location of the vehicle and effectively track the vehicle in real time.In this paper,the traditional vehicle detection algorithm is analyzed.Aiming at its shortcomings,a vehicle detector based on multi-scale feature graph prediction SSD is designed by using deep learning technology.The detection results are combined with the improved Camshift tracking algorithm with Kalman filtering algorithm to achieve automatic vehicle tracking.The specific work of this paper can be summarized as follows:1.The advantages and disadvantages of the inter-frame difference method,the improved gaussian mixture model(MOG2)and the ViBe algorithm were analyzed and discussed.The comparative experiment showed that the ViBe algorithm was selected as the moving vehicle detection algorithm to achieve the optimal effect.In view of the problem of fixed threshold and "ghost" interference in the traditional ViBe algorithm,the OTSU inter-class variance method is adopted to overcome the problem of fixed value,and the inter-frame difference method is added to improve the "ghost" interference.Comparative experiments show that the improved ViBe algorithm based on the inter-frame difference method not only overcomes the fixed value problem,but also can quickly eliminate the influence of "ghost" on the detection results,and has a good detection effect.2.Aiming at the problems in traditional vehicle detection algorithm,this paper designed a based on the multi-scale feature map to predict the SSD vehicle detector,its basic idea is: to the transmission network(CNN)before use,produce a series of fixed size boundary box set and the target category box score,and then through the maximum suppression algorithm for final testing;The experimental results show that ssd-based vehicle detector has a good detection effect in both recognition performance and time efficiency.3.Vehicle detection and tracking are realized by using multi-scale feature graph prediction SSD vehicle detector combined with Camshift tracking and Kalman filtering algorithm.It mainly includes motion detection module,vehicle detection module and vehicle tracking module.Vehicle detection and tracking first,the improved ViBe algorithm was used to separate the background before and after video,and the motion area in video was extracted.Then,ssd-based vehicle detector was used to detect the extracted motion area.Finally,Camshift tracking and Kalman filtering algorithm were combined to realize real-time tracking of vehicles.