Application Research Of Traffic Signs Detection And Recognition Technology Based On Deep Learning

The popularity of cars has put pressure on traffic management while facilitating people’s travel.People have been trying to use modern science and technology to find ways to improve the safety of vehicles while driving.In recent years,deep learning has achieved good results in the field of industrial vision,and has been widely used in assisted driving systems and automatic driving systems.The computer vision-based traffic sign detection and recognition function is an indispensable part of the assisted driving system and the automatic driving system.Based on the review and analysis of the basic issues,this paper mainly uses the deep learning technology to study the traffic sign detection and identification.The main research work and contents are as follows:(1)Aiming at the low recognition efficiency of traditional traffic sign matching recognition algorithm,based on SIFT feature matching algorithm,a SIFT feature matching algorithm based on multi-block matrix is proposed.Experiments verify that the algorithm is better than the classical SIFT feature matching algorithm.Higher rates and better real-time performance.(2)Large-scale available marker data is one of the key factors for the success of deep learning models in computer vision problems.In order to ensure the smooth training and verification of the deep learning model,2777 actual traffic scene traffic sign images were manually collected,and then the image processing technology was used to enhance the data,and finally a traffic sign data set containing 24993 images was generated.(3)The traffic sign recognition method for traditional deep learning has high data requirements and low learning efficiency in small sample data learning.This paper uses migration learning technology to improve it,so that the improved algorithm is under micro-sample conditions.The recognition rate has been significantly improved.(4)Since the detection task and the recognition task are separated in the traditional traffic sign recognition system,the whole recognition process is complicated and thereal-time and robustness are poor.To solve this problem,this paper uses a deep learning algorithm based on the combination of traffic sign detection task and recognition task.(5)Aiming at the shortcomings of the current multi-task joint detection and recognition algorithm for large target detection error,this paper introduces the residual network structure based on the YOLOv2 algorithm and uses the multi-scale feature map to generate small resolution targets.Feature semantics.Through experimental comparison,it is verified that the improved algorithm improves the detection and recognition accuracy of small-resolution traffic signs based on the excellent real-time and robustness of YOLOv2.