Research On Traffic Sign Recognition Method Based On Deep Learning

Nowadays,with the rapid development of economy and society,cars provide us with a lot of conveniences.Automatic driving and unmanned driving are also stepping into the commercial field.At present,there are some problems in traffic sign detection,such as missing detection,low recognition accuracy and high model complexity.How to recognize traffic signs quickly and accurately is an urgent problem.Based on the above problems,this paper proposes a traffic sign detection model based on BYOLOV5+ECA,which has practical value in real-time traffic sign detection.The main work of this paper is as follows:Firstly,in view of the unbalanced distribution of categories in TT100 K dataset,it is necessary to filter the TT100 K dataset,and separate the samples with more than100 signs and information to produce traffic signs containing 45 categories.The dataset images are uniformly processed to 416 * 416 in size and are divided into train and test folders at a 7: 3 scale.Secondly,in view of the fact that YOLOv5 s is a lightweight network and traffic signs are small targets,this paper proposes an improved YOLOv5 s algorithm with ECA module to improve the feature extraction ability of convolution layer.Improved recognition rate,speed per image about 0.162 s.Finally,in view of the fact that traffic signs are small targets with a small proportion,a clustering algorithm is adopted to find the a priori box size in line with the characteristics of the targets,and the feature extraction network layer is replaced by Bi FPN,the original simple two-way fusion is changed into complex two-way fusion features,and weight information is introduced to better balance the characteristics information at different scales.Based on the above main work,experimental analysis is carried out.Experimental results show that the proposed improved YOLOv5 network model has higher recognition precision and accuracy,reduces the missing rate of traffic signs and improves the ability of convolution layer feature extraction.The overall experimental design and experimental results can meet the needs of real scene traffic signs real-time detection.