Research On Traffic Sign Recognition Technology In Driving Assistant System

With the development of the national economy,cars are gradually invading all households,and traffic safety issues are also a hotspot for research.With the advent of the driving support system,the damage caused by traffic safety problems can be greatly reduced,not only to assist the driver in driving,but also to avoid the traffic safety problems caused by the driver’s carelessness as much as possible.By giving consideration to the driver when parking,the burden on the driver while driving is greatly reduced and safety is improved.As a key technology in driving assistance systems,road sign recognition technology allows drivers to prepare in advance because they can remind them of traffic signs in front of the road.This paper addresses the problems of inadequate accuracy,slow speed,and large models of traffic sign recognition algorithms.This article mainly completes the following:Improved pre-processing steps in traditional traffic sign detection and recognition methods,combined with gamma correction and contrast limiting adaptive histogram algorithms to reduce the impact of lighting factors on traffic sign recognition,for traffic sign detection and recognition.It will be realized.Day and night.We analyzed the algorithms of the YOLO series based on deep learning,and finally selected the YOLOv5 s algorithm as the template algorithm for the follow-up work in this paper.The coordinate loss function and non-maximum suppression algorithm of the YOLOv5 s algorithm have been improved.First,improve the coordinate loss function of the YOLOv5 s algorithm,use the EIOU loss function to replace the GIOU loss function used in the YOLOv5 s algorithm,optimize the training model,improve the accuracy of the algorithm,and speed up target recognition.To realize.Weighted cluster NMS to improve the use of YOLOv5 s itself.The weighted NMS algorithm can improve the accuracy of detection frame generation.Assuming the problem of large traffic sign detection algorithm model and slow detection speed,we will compare the lightweight neural networks Mobile Netv1,Mobile Netv2 and Ghost Net,and finally select the lightweight network Ghost Net to improve the network structure of the YOLOv5 s algorithm.With the addition of the attention mechanism CBAM module,the YOLOv5s-GN-CBAM network is designed to reduce the weight of the algorithm without reducing the accuracy of the algorithm as much as possible.Ghost Net is very suitable for mobile use,so the improved algorithm is suitable for practical applications such as driving assistance systems.Also,through Python’s pyqt library,traffic sign detection and recognition systems will now more intuitively display the detection effects of algorithms on traffic signs,simulating traffic sign detection and recognition in real-world scenes.It is designed.In the experiment,we used the traffic sign data set of CCTSDB,and for the data set used this time,we used the image of the traffic sign taken by ourselves.Experimental results show that the m AP value of the model trained by the improved YOLOv5 s algorithm reaches 84.35%,which is 6.23% higher than the original YOLOv5 s algorithm.In addition,experiments have shown that the YOLOv5s-GN-CBAM network reduces the size and amount of parameters of the training model without compromising detection accuracy.It is suitable for use on mobile devices such as driving assistance systems.