Research On Traffic Sign Detection Technology Based On YOLOv3 Network

In real scenes,motor vehicles often have a higher driving speed.In order to detect the traffic signs accurately and in time,when the vehicle is far away from the traffic signs,the on-board camera must collect images of the traffic signs on the road,so that the traffic sign detection system has time to analyze and process the traffic signs on the road.And timely feedback to the driver.Long-distance image collection will reduce the proportion of traffic signs in the detected images,which will result in fewer features related to traffic signs in the detected images,and ultimately lead to a decline in the performance of the traffic sign detection system.In order to solve the above problems,this article is based on the YOLOv3 network,which is good at small target detection and has good real-time performance,to further improve the network’s detection performance of traffic signs.First of all,in order to solve the problem that the proportion of traffic signs in the detected images is small and difficult to detect,this paper proposes a spatial pyramid pooling YOLOv3(Spatial Pyramid Pooling YOLOv3,SPP-YOLOv3)traffic sign detection network.The network introduces the spatial pyramid pooling(Spatial Pyramid Pooling,SPP)module after the third residual block,the fourth residual block,and the fifth residual block of the backbone network respectively,and integrates local and global features to enhance the expressive ability of features,thereby improving the accuracy of the network’s detection of traffic signs.Secondly,the Feature Pyramid Network(Feature Pyramid Network,FPN)structure of the SPP-YOLOv3 traffic sign detection network has inconsistencies between different scales,that is,in the process of back propagation,the gradient calculation includes both positive and negative samples.This inconsistency will cause certain interference to the gradient calculation,reduce the effectiveness of the feature pyramid network structure,and then affect the accuracy of the network’s detection of traffic signs.In order to solve this problem,this paper proposes Adaptively Spatial Feature Fusion(Adaptively Spatial Feature Fusion,ASFF)and Spatial Pyramid Pooling YOLOv3(Adaptively Spatial Feature Fusion & Spatial Pyramid Pooling YOLOv3,ASFF&SPP-YOLOv3)traffic sign detection algorithms.This algorithm suppresses the inconsistency between different feature scales by learning how to filter conflict information in space,and improves the scale invariance of features.In turn,the network’s detection performance of traffic signs is improved.