Improvement And System Implementation Of Traffic Sign Recognition Model In Complex Scenarios Based On YOLOv5

With the introduction and research of autonomous driving,traffic sign recognition has become a hot topic for many scholars.As a key technology for autonomous driving and assisted driving,studying traffic sign recognition technology can optimize the driver’s driving experience,alleviate traffic pressure,and reduce traffic accidents.In actual traffic scenarios,there are complex traffic conditions such as weather changes,signs that are old,faded,or obstructed,and signs that are near,far,and small during driving,resulting in inaccurate model recognition and high missed detection rates.To address these issues,this article proposes improvement strategies for the dataset and network model,and verifies the effectiveness through comparative experiments.First of all,in the data set part,the original data set lacks data on complex road conditions such as foggy days,dark nights,and signs being blocked,and there is a serious problem of sample imbalance.This paper simulates the road conditions of complex field scenes such as brightness,chroma change,and sharpening processing for the original data set and is used to expand the data set.After balancing,the number of each category is kept between 500 and 550,ensuring the balance of data samples,After expanding the dataset,the model in validation set 1 map@0.5 and map@0.5 0.95 increased by 7.02%and 6.02%,respectively,in validation set 2 map@0.5 and map@0.5 The results of 0.95 increased by 15.5%and 12.9%,respectively.Secondly,in the network structure section,in order to address the issues of low accuracy and high miss rate in small target recognition,this paper adds a small target detection layer to the YOLOv5 model,which preserves more shallow semantic information of small targets in high-level features and reduces the loss of important features;Then,by using the kmeans clustering algorithm,the preset anchor box size in the model is modified to better match the dataset in this paper;Finally,attention modules were added to Backbone and Neck respectively to improve the sensitivity of the model to the recognition area.The experimental results indicate that the improved model performs well on validation set 1 map@0.5 and map@0.5 0.95 increased by 5.72%and 4.56%respectively on validation set 2 map@0.5 and map@0.5 0.95 increased by 5.4%and 3.7%respectively,and the size of the model was 16.1MB,an increase of 2.1MB compared to the original model.This article proposes an improvement strategy that effectively improves the recognition accuracy of the model under complex road conditions while ensuring its real-time performance.Finally,this article will conduct comparative experiments with other object detection models to demonstrate the superiority of the proposed model.And the real-time recognition system of traffic signs is developed.The main functions include Model selection,picture recognition,video recognition,calling camera recognition,and saving the recognition results.