Research On Traffic Sign Recognition Algorithm Based On YOLOv4

With the rapid development of China’s economy,the number of domestic motor vehicles has increased significantly,and the frequent occurrence of traffic safety problems has attracted extensive attention.However,with the continuous development and improvement of Intelligent Vehicle and Intelligent Transportation System in recent years,the number of traffic accidents began to decrease.Traffic sign detection is a very important part of Intelligent Vehicle and Intelligent Transportation System.It is mainly used in driving assistance system,automatic driving system and traffic sign maintenance.The vehicle camera senses the environment,obtains the environmental information around the vehicle,and sends the obtained traffic signal information or traffic sign information to the traffic sign detection system,so as to enable the driver or vehicle to make correct analysis and judgment,ensure the smooth and safe driving of the vehicle.Therefore,how to detect traffic signs accurately and quickly still has very important theoretical and practical significance.Through the comparison of data sets at home and abroad,this paper selects TT100K data set as the experimental data set.Aiming at the problems of insufficient sign types and lack of signs under complex weather conditions in TT100K data set,the type information of guidance,tourist area,road construction safety and auxiliary signs missing in the data set is added by means of sign replacement.Through the way of data enhancement,the complex environmental information of rainy days,snow days,haze and different lighting conditions lacking in the data set is added.By means of sign replacement and data enhancement,a total of 65 types and 15002 pieces of traffic sign data information for experiment are obtained.After that,the traffic sign data is labeled by Labelimg labeling tool,and the labeled data set is divided into training set,verification set and test set,then the preparation of the experimental data set is completed.The traffic sign detection algorithm based on deep learning has higher detection accuracy and speed.By comparing the current mainstream target detection algorithms,this paper selects YOLOv4 algorithm as the basic algorithm of the experiment.Aiming at the low accuracy of YOLOv4 algorithm in detecting small target objects and target objects in complex environment,K-means plus plus clustering algorithm is used to re-cluster the a priori frame of the data set to obtain the a priori frame size suitable for the data set.By adding the coordinated attention mechanism module,the ability of the trunk module to locate and detect the target is enhanced.By improving the feature fusion module and adding new convolution layer and feature compression layer,the detection ability of the network module for small target objects and objects in complex environment is improved.Through experimental analysis and verification,the m AP of the improved YOLOv4 algorithm is increased by 6.95%.In order to apply the improved YOLOv4 algorithm model to the traffic sign detection system,this paper completes the traffic sign detection system based on PyQt5 graphic program framework through system demand analysis and system function design.The traffic sign detection system can preprocess fuzzy images,haze images and images with poor light conditions in the evening,and detect the traffic sign information in single picture,multiple pictures,cached video and real-time video.After testing,the system can complete the above functions well.