Research On Vehicle Detection Method Based On Deep Learning In Traffic Safety

With the continuous advancement of urbanization and the increase in the number of road vehicles,it is an important task of intelligent transportation system management to take photos of moving vehicles through electronic equipment,and to detect and identify them.At the same time,it is of great significance for improving and solving road traffic safety problems.For vehicle detection,how to improve the problems of missed detection,false detection and repeated detection in vehicle detection,and how to optimize the design of lightweight network models are still facing serious challenges.Through the research of target detection algorithms,proposes two improved vehicle detection methods.One is based on the improved YOLOv3 vehicle detection algorithm,which uses the K-means clustering algorithm to determine the number of target candidate frames and the dimension of the aspect ratio according to the inherent width and height characteristics of the vehicle,then reset the model parameters according to the results of clustering,so that the network model has a certain pertinence in vehicle detection.At the same time,feature splicing and residual mapping operations are added to strengthen the information fusion between feature maps and increase the training accuracy of the model,improved the problems of missed detection,false detection and repeated detection.The other is based on an improved SSD.The algorithm performs feature scaling,feature mapping and deep feature fusion operations on feature maps of different scales,and uses deep separable convolution in residual mapping to reduce the parameter of the model.At the same time,the regression loss function is optimized to reduce the parameter amount of the model while ensuring the detection accuracy,and improve the detection speed of the model.