Research On Vehicle Detection And Recognition Technology Based On Vision

Vehicle detection,identification and tracking based on video image is a core technology in the field of intelligent transportation system,which has important scientific research significance.The robustness,high-scale insensitivity,real-time and high-precision of vehicle detection and identification algorithm based on video image is an important topic studied by scholars.Therefore,this paper studies the problems of low accuracy and slow detection speed of vehicle recognition and detection algorithm based on video image.The main research contents are as follows:1.It expounds the difficulties faced by vehicle detection and recognition in urban roads and its importance in intelligent transportation system.On the basis of summarizing the current research status of vehicle detection and recognition technology,and it summarizes the advantages and disadvantages of the current vehicle detection and recognition methods.2.Aiming at the problem of low-quality video image,based on the analysis of background difference method,inter frame difference method and edge detection method,an image preprocessing method integrating background difference method,inter frame difference method and edge detection method is proposed.The method uses the above three algorithms to process the vehicle target area respectively,so as to obtain three different foreground images respectively.Then,the decision level fusion of the obtained three foreground images is carried out by using evidence theory to obtain a more complete foreground image of the vehicle information.This method makes full use of the advantages of the three algorithms to obtain better target processing images,which lays a solid foundation for achieving better vehicle detection and recognition effects in the future.3.Aiming at the problems of low accuracy and slow detection speed in vehicle detection and recognition algorithm,an improved algorithm based on YOLOv5 is proposed.The algorithm introduces the attention mechanism module to act on each detection layer of YOLOv5 model,so that in the process of feature extraction,the network can enhance effective features and weaken useless features in channel domain and spatial domain.In order to reduce the loss of feature information caused by too many network layers,a weighted bidirectional feature pyramid network is added to fuse the features of the network.At the same time,in order to improve the accuracy of detection,CIo U loss function is used to replace the original loss function,the distance between the two frames is taken as the penalty term of the loss function,and the relative proportion of rectangular frames is further considered.Combined with Deep Sort,vehicle tracking is carried out to optimize the network model and improve the detection effect.Finally,the target detection public data set PASCAL VOC and the field shot data set are selected to carry out multiple control experiments on the improved YOLOv5 algorithm.Through the comparison with the detection effect of the original YOLOv5 model,the results show that the improved model effectively improves the accuracy and speed of target detection and recognition.Aiming at the improvement research of the above vehicle detection and identification algorithm,the moving vehicle detection and tracking system is preliminarily designed and completed.The practice shows that the technical scheme proposed in this paper is feasible and effective,and meets the technical requirements of image-based vehicle detection and recognition in most cases.