Research On Traffic Sign Recognition And Ranging Method Based On Deep Learning

Today,with the rapid development of computer vision technology,the auxiliary system of intelligent vehicle is constantly updated iteratively.Among them,the perception system of road environmental conditions has been greatly developed.The auxiliary system of intelligent vehicle can timely and accurately perceive the complex road environment around the vehicle body.This function is of great research significance to avoid mutual collision between vehicles and ensure daily driving safety.In view of the complexity of the actual road background,this paper studies the methods of traffic sign detection and recognition,and uses monocular vision algorithm for ranging to help the driving system measure the traffic signs in real time,so as to assist in judging the real-time road conditions.The main work of this paper is as follows:Aiming at the problems of missing detection of small targets and poor detection effect under weak light conditions,a traffic sign recognition system based on weighted two-way yolov5s model is designed.Firstly,the weighted bidirectional feature pyramid network(bifpn)network structure is used to replace the feature pyramid module in the feature fusion module of yolov5s model,so as to realize the two-way cross-scale connection and weight the feature fusion module,improve the detection speed and detection rate,and use the se channel attention mechanism to strengthen the region of interest,so as to solve the problem of missing detection of small targets;Then,the brightness and contrast of the unified image are adjusted by fast gamma correction.After normalization,pre compensation and anti normalization,the image to be detected is enhanced,so as to improve the problem of poor detection effect under weak light conditions.Finally,the data set cctsdb is selected to establish the training set and test set of the detection system,and the experimental results show that it is consistent with the improved model affb based on yolov5s＿Compared with the recognition system of yolov5s,the recognition frame rate of the recognition system designed in this paper is 73 FPS,and the recognition speed is improved by 12 FPS.The map of traffic sign recognition target is77.9%,an increase of 4.1%;The map for the target recognition of traffic sign test set under weak light conditions is 70.8%,which is 7.6% higher than the recognition result without gamma correction.In order to improve the perception ability of the unmanned vehicle assistance system to the distance of traffic signs,the monocular visual ranging method based on color or shape is improved,so that the unmanned vehicle can make predictions based on the distance more accurately,take braking or measures to slow down.Firstly,the traffic signs are detected and recognized by the improved weighted two-way YOLOv5s model,then the recognized traffic signs are intercepted,and the edge of the canny operator is extracted,and binarized to obtain more than four feature points,and finally use PnP The algorithm obtains the distance estimation value of the traffic sign by calibrating the camera parameters.The experimental results show that the improved ranging accuracy is improved compared to the monocular vision ranging method based on color and shape.By measuring the distance of the traffic sign targets within 3-16 meters,when the distance between the car and the recognition target is 16 meters,the absolute error is 0.32 meters,and the relative error is 2.04%,and its accuracy meets the actual needs.