Research On Detection And Identification Of Driverless Traffic Safety Signs Based On Deep Learning

The car is an indispensable means of transportation in today’s society.Road safety has also become a hot topic in today’s society.People increasingly rely on and demand for safe and convenient Intelligent Traffic System(ITS).As a key issue in ITS System,the accuracy of Traffic safety sign detection and identification plays a vital role in the safe operation of the whole System.People have been trying to solve road traffic safety problems with the help of modern technology.In recent years,deep learning technology has been very popular and widely applied in the field of unmanned driving.In this paper,the improved deep learning algorithm is applied to the detection and recognition of traffic safety sign data set,and the main research contents are as follows:(1)A large number of data sets are used as a prerequisite for training a robust model.In this paper,part of the open source TT100 K traffic safety sign data set is adopted in China.On this basis,contrast transformation,Gaussian transformation,morphological operation and other data enhancement techniques are used to solve the problem of unbalanced and insufficient data distribution.(2)Based on the low accuracy of the traditional feature detection algorithm for traffic safety signs,this paper proposes a method to improve the traditional Hough circle detection R-Hough algorithm by using the principle of gradient algorithm and TRUNC binarization algorithm,The measurement accuracy was increased by 14.8% compared with that without improvement.(3)At present,the Learning efficiency of the convolutional neural network model is low in a small number of samples.In this paper,a fusion Transfer Learning algorithm is proposed to improve the network structure of the CNN-RESNET convolutional neural network model.In the case of a small number of samples,recognition rate of the improved algorithm is 7%higher than that of the unimproved algorithm.(4)In order to improve the real-time performance of detection and recognition,YOLOV4’s multi-task cascading detection and recognition algorithm was used.YOLOV4 shall sex difference and the small target detection effect is poor,in this paper,from the data,vector optimization strategy and network structure three aspects to improve the original YOLOV4,the improved model in small target detection,target and real time relative to the original YOLOV4 algorithm is improved,and provides sufficient time for unmanned vehicles to make behavioral decision,ensure the security.