Research On Algorithm Of Traffic Sign Detection And Recognition In Complex Environment

In recent years,with the rise of deep learning,traffic sign detection has made great progress.Due to the complex and changeable traffic environment,it is still a challenging problem to detect small traffic signs in the actual scene.Therefore,based on YOLOv5,this paper detects and recognizes different types of traffic signs in complex environment.The main research contents are as follows:(1)This paper presents a bidirectional feature fusion algorithm based on YOLOv5s6.Considering that the input features of different scales have different contributions to the output features,this paper adds an additional weight to the input features of different scales to let the network learn the importance of each input feature,so as to improve the multi-scale detection ability of the network.Verified by tt100 k traffic sign data set,the mean Average Precision of the improved algorithm is increased by 2.4%,and the recall rate of the network is increased by 2.3%.By comparing the detection effect graphs,the improved algorithm can reduce the frequency of missed detection and false detection,and improve the confidence of some detected targets.(2)Two traffic sign detection and recognition algorithms based on attention mechanism are proposed in this paper.Aiming at the problems of blurred and occluded traffic signs in images,the C3-MHSA module is first designed to model the context of the global information of the feature map,and obtain the dependence between long-distance features,so as to enhance the attention of the model to the features of occluded targets.In addition,by introducing the orientation and location attention modules,the feature maps are encoded as orientation-aware and location-aware attention feature maps to enhance the representation of objects of interest.Through the verification of TT100 K traffic sign data set,the m AP value of improved method 1 is increased by 1% compared with the original model,the m AP value of improved method 2 is increased by 1.81% compared with the original model,and the Recall value is increased by 1.38%.After the comparison of the detection renderings,the two improved methods can effectively improve the detection effect of blurred and occluded objects.(3)This paper proposes a lightweight traffic sign detection and recognition model: the M3-YOLO model.By combining the lightweight convolution network Mobile Net V3 with YOLOv5,the complexity of YOLOv5 model is reduced,the speed of traffic sign detection is improved,and the requirements of the model for the hardware configuration of automatic driving environment perception are reduced.In addition,the Coordinated Attention module is introduced to enhance the extraction ability of the model for the effective information of traffic signs.Through the experimental verification,the m AP value of the improved model is 78.6%,and the Recall value is 70.3%.It can be seen that the model can still maintain a good detection effect on traffic signs while reducing the amount of parameters and calculation,and reduce the hardware cost to meet the needs of deployment on a low computing platform.