Research On Road Traffic Sign Detection Method Based On Deep Learning

Traffic sign detection is an important part of intelligent driving.An efficient and accurate traffic sign detection system can effectively guide drivers to regulate driving and reduce traffic accidents.Traffic sign detection is easily affected by light intensity,extreme weather and distance.In order to meet the actual needs,the traffic sign detection system needs to better balance detection speed and detection accuracy.This thesis investigates road traffic sign detection methods based on deep learning.The main elements of the research in this paper are as follows:Aiming at the problem of low accuracy and slow recognition speed of YOLOv3 for small target traffic sign recognition,this thesis proposes a traffic sign detection algorithm based on YOLOv3-P-4L network.The network pruning method is used to compress the Darknet-53 backbone network,which reduces the calculation amount of the network model and improves the running speed of the model;using a four-scale detection structure,the network makes full use of the shallow feature information of small traffic signs to improve the performance of small target traffic sign detection;using K-means++ algorithm to cluster and analyze the bounding box size of the targets in TT100 K traffic sign dataset to generate a priori boxes with more suitable size and improve the detection rate of a priori boxes;Finally,the loss function is optimized to increase the penalty of classification prediction,and the confidence loss and classification loss of each scale prediction are multiplied by the corresponding weights to improve the accuracy of traffic sign classification.The test results based on the TT100 K traffic sign dataset show that the accuracy of the YOLOv3-P-4L algorithm is92.86%,which is 2.30% better than the YOLOv3 model,and the model size is reduced by 9.70%.Aiming at the problem of low detection accuracy of YOLOv3-P-4L algorithm in complex environments such as haze weather and low light intensity,this thesis proposes a traffic sign detection algorithm based on M-YOLO-FSC network in complex scenes.In this thesis,the backbone network Darknet-53 of YOLOv3 is replaced by the lightweight network Mobile Netv3 to improve the model detection speed;using the Focus module,image pre-processing is performed at the front-end of the Mobile Netv3 backbone network to avoid feature information loss and insufficient shallow feature extraction;introducing a spatial pyramid pooling network to prevent image distortion;introducing a cross stage partial network to eliminate repetitive features in the computation process,The introduction of cross stage partial networks eliminates repetitive features in the computation process,reduces computational bottlenecks,and ensures detection accuracy while reducing computational effort.The test results based on CCTSDB2021 complex scene traffic sign dataset show that the average detection accuracy of the algorithm is 93.6%,which is 2.30% higher than YOLOv5l;the average detection time per image is 84.2ms.This model has a good balance between detection accuracy and detection speed.Based on the above algorithm,this thesis designs and implements a traffic sign detection system.After system testing,the results show that the traffic sign detection system has perfect functions and good performance.This thesis includes 44 figures,14 tables,and 80 references.