Key Technology Of Full-Element Traffic Sign Detection And Tracking Research And Implementation

Full-element traffic sign detection and recognition and tracking is a key element of intelligent transportation system.In view of the different types of traffic signs in the task of traffic sign detection and recognition and tracking,the wide variety of traffic signs,the efficiency of detection and recognition and the low degree of automation,this paper focuses on the task of detection and recognition and tracking of different types of complex traffic signs,and conducts research on the all-element traffic sign detection,tracking and text recognition models respectively,and the main work accomplished includes:(1)A YOLOV5-based traffic sign detection and recognition model is proposed.To address the current problems of low accuracy and reliability in road traffic sign detection and recognition,this paper explores the multi-scale,multi-category features of graphic traffic signs based on YOLOV5,and makes YOLOV5 detection algorithm by incorporating effective channel attention module(ECA),adaptive spatial feature fusion network(ASFF)and SIOU bounding box loss function in the backbone network.improvements.The experimental results show that the m AP of the improved traffic sign detection and recognition model reaches92.8%,which improves the accuracy of traffic sign detection and recognition by 2.3%.(2)A traffic sign detection and tracking model combined with Byte Track is established.Since traffic sign recognition and detection systems are mostly used in the process of driving driving,in order to solve the problems of poor tracking efficiency and easy loss of targets of current traffic sign detection and recognition algorithms,this paper uses the improved YOLOV5 model proposed in Chapter 1 as a detector and Byte Track model as a tracker,and uses the data association algorithm of the recall mechanism to reduce the impact of low confidence detection frames on the target predicted trajectory and improve it by introducing the noise-adaptive Kalman filter algorithm.The experimental results show that the MOTCA and MOTA of the improved traffic sign detection tracking model are 89.6% and 74.3%,respectively,and the IDs are reduced to 20,which effectively improves the accuracy and robustness of traffic sign tracking and reduces the number of ID jumps of the obscured targets.(3)A CRNN-based semantic recognition model for traffic signs is proposed.For the current research in the field of traffic sign detection and recognition mostly focuses on graphic traffic signs,the paper detects text-based traffic signs with the improved YOLOV5 network,and improves the CRNN semantic recognition algorithm by text direction correction and introducing Dense-Net dense convolutional network.The experimental results show that the m AP of text-based traffic sign detection using the improved YOLOV5 network can reach88.6%,and the recognition accuracy and processing time of the improved CRNN traffic sign recognition model can reach 87.21% and 17.1ms respectively,which is 6.61% and 2.5ms higher than the basic CRNN model,effectively reducing the probability of misrecognition and improving the recognition accuracy while reducing the computational effort.The calculation volume is reduced and the recognition accuracy is improved at the same time.The paper integrates the above research to build a system that can detect and track allelement traffic signs,and has been tested to confirm its reliability,real-time and stability,and to show the practical significance and feasibility of the above research in the field of detection and tracking of all-element traffic signs and intelligent transportation systems.