Based On The YOLOv5s Traffic Sign Identification Research

Traffic sign recognition is the basis of automatic driving technology and the premise of fully automatic driving.However,the existing traffic sign recognition algorithms still have some problems in recognition efficiency and detection accuracy.In order to meet the complex traffic sign recognition requirements of self-driving,this paper proposes a traffic sign recognition algorithm based on the improved YOLOv5 s,aiming at realizing high-precision and high-efficiency traffic sign recognition.The details are as follows:(1)Firstly,the technical scheme to realize high-precision and high-efficiency traffic sign recognition is determined.In this paper,by analyzing the current research status of traffic sign recognition in China,and comparing the performance of the classic Fast R-CNN,SDD and YOLO series algorithms in traffic sign recognition,combined with the principles of various algorithms,it is found that YOLOv5 s algorithm performs well in comprehensive performance such as recognition accuracy and recognition speed,and the technical scheme of realizing traffic sign recognition based on YOLOv5 s is determined.(2)Secondly,aiming at the problem of lack of traffic sign labeling data,this paper uses transfer learning method to enhance the accuracy of traffic sign recognition.Although there are a lot of traffic sign data in the network,they are often not labeled,and manual labeling is time-consuming and laborious.The model pre-trained in Imagenet is used for fine-tuning to improve the recognition accuracy under limited labeled data.(3)Subsequently,Ghost-Conv is embedded in YOLOv5 s,which can significantly improve the calculation and processing efficiency without losing accuracy.The feature map is generated by the convolution kernel size of half of the original convolution,and then the feature map is processed by cheap calculation to generate another part of the feature map.Finally,the two parts of the feature map are spliced into a complete feature map by Concat operation,which effectively reduces the computational complexity of YOLOv5 s,improves the computational processing efficiency,and keeps the recognition accuracy unchanged.(4)Finally,the design and model verification analysis of traffic sign recognition system based on YOLOv5 s model are completed.According to the characteristics of traffic identification,the whole system framework is designed,including hardware selection and software implementation.Through the comparative analysis of experiments,the model proposed in this paper performs better on the loss curve and PR curve.By adding two SE modules,the network model obtains better performance on the PR curve without affecting the training process.The improved model is faster than the original YOLOv5 s,and can meet the real-time detection requirements of 30 frames per second(FPS).Through the above optimization measures,the traffic sign recognition algorithm based on the improved YOLOv5 s proposed in this paper can realize high-precision and high-efficiency traffic sign recognition,which lays a solid foundation for the application of autonomous driving technology.