Based On The Application Of Improving YOLOv4 Algorithm In Road Traffic Sign Recognition

For the traffic sign recognition system,its core role is to accurately and timely identify the road traffic sign information to obtain the current road conditions and driving environment,thereby reminding and assisting the driver to control the road information and correct the wrong traffic behavior.Traditional target detection algorithms are susceptible to many factors,which lead to problems such as difficulty in algorithm implementation,low recognition accuracy,and slow recognition rate.Therefore,this thesis chooses the best network model YOLOv4 network to realize the detection and recognition of traffic signs.Because the YOLOv4 network model generates too much redundant information when detecting and recognizing images,it affects the detection accuracy of the model.Based on this problem,this thesis proposes a pooled structure SAP module that integrates the attention mechanism to perform the YOLOv4 network Optimize to solve this problem.As a traffic sign recognition system,it is generally used in high-speed cars,so the accuracy and real-time requirements of recognition are high.For the YOLOv4 network and the improved SAP-YOLO network,the models are too large,which makes it difficult to be applied in the embedded traffic sign recognition system.Therefore,this article uses the lightweight neural network Mobilenet network to optimize the SAP-YOLO network.It can reduce the model size and increase the detection rate while maintaining the detection accuracy.The main work of this thesis is as follows:(1)Due to the lack of domestic traffic sign data sets,this thesis selects a self-made traffic sign data set to train and test the improved network,and annotates and enhances the data set.(2)Aiming at the problem of excessive redundant information generated by the YOLOv4 network when detecting and recognizing images,the SAP pooling structure is proposed to solve the problem of information redundancy in the multi-channel pooling layer,and by introducing non-maximum values The suppression algorithm removes redundant prediction frames,thereby improving the detection accuracy of the algorithm.(3)Re-clustering the a priori boxes of the self-made data set through the K-means++algorithm to generate 9 new a priori boxes.(4)Aiming at the problem that the YOLOv4 network parameter is too large and the detection rate is slow,the model size and the detection rate are increased by fusing the lightweight neural network Mobilenet.Finally,the improved algorithm M-SAP-YOLO network is trained and experimented.Experimental results show that the algorithm proposed in this thesis has a good recognition effect on various traffic signs,with an average recognition accuracy of91.99%.The model size of the improved algorithm is only 58.97 MB,which is only onefifth of the YOLOv4 network.First,the detection rate has reached 44 frames per second,which is nearly 90% higher than the detection rate of the YOLOv4 network,indicating that the improved algorithm M-SAP-YOLO in this thesis has a certain degree of robustness.