Research On Lightweight Traffic Sign Recognition Model Based On Coordinate Attention

In recent years,all countries in the world have focused on the research and promotion of new energy vehicle technology,and new energy vehicles with auxiliary driving system and auto drive system are gradually popularized.Traffic sign recognition technology is a research hotspot in the application of visual navigation and computer vision in intelligent driving.In the task of road traffic sign recognition,minor mistakes may bring disastrous consequences.This thesis analyzes the current research status of traffic sign recognition algorithms at home and abroad,and proposes a lightweight traffic sign recognition model based on coordinate attention(CA),aiming at the problems of unbalanced detection speed and recognition accuracy of existing traffic sign recognition models,as well as the difficulty of detecting occluded targets and small targets.The main research work is as follows:(1)To solve the problem that it is difficult to detect occluded targets and small targets,YOLOv5(You Only Look Once version 5)has improved and a traffic sign recognition model based on CA is proposed.Firstly,CA mechanism has integrated into the backbone to effectively capture the relationship between location information and channels,so as to obtain the region of interest more accurately,while avoiding excessive computational overhead;Then,a crosslayer connection is added to the feature pyramid network(FPN)to fuse more feature information and improve the feature extraction ability of the network without increasing the cost,so as to improve the detection effect of occluded targets;Finally,the improved CIo U(Complete Intersection over Union)function has introduced to calculate the regression loss,alleviate the uneven distribution of sample size in the detection process,and further improve the recognition accuracy of small targets.Applying this model on chinese traffic sign dataset TT100K(Tsinghua-Tencent 100K),the recognition accuracy reaches 91.5%,the recall reaches86.64%,which is 20.96% and 11.62% higher than the traditional YOLOv5 n model,and the detection speed reaches 140.84 FPS.(2)In order to improve the detection speed of the network,the above model was lightweight and optimized,and a more balanced lightweight traffic sign recognition model has further proposed.The lightweight structure of C3 module has designed and the coordinate attention mechanism was integrated to improve the feature extraction ability of the network.The backbone and neck were pruned,and redundant network layers were removed to further reduce the parameters.For the public traffic sign data set,the label assignment strategy and loss function are improved to alleviate the imbalance of positive and negative samples and better calculate the loss,so as to make the training more stable and efficient.The experimental results show that this method better realizes the tradeoff between speed and accuracy.Only 0.85 M parameters are used to achieve 91.9% recognition accuracy,and the frame processing rate reaches 360 FPS,which is 21.5% and 26.67 FPS higher than the traditional YOLOv5 n.