Research On Traffic Sign Detection In Unmanned Driving Based On Neural Network

With the improvement of people’s living standards,motor vehicles are appearing more and more frequently in people’s daily lives.Not only as an indispensable means of travel for people,but also begin to enjoy the fun of driving motor vehicles and pursue driving Experience.However,the threat of traffic accidents to people’s lives and property can not be ignored,so more and more people are looking for ways to reduce the probability of traffic accidents.Compared to fatigue-prone humans,computers stand out by virtue of their powerful computing power and stable logical judgment.Intelligent Transportation System(ITS)has attracted a lot of attention and energy from academia and automobile manufacturers in recent years.As an important part of intelligent transportation systems,unmanned driving and driver assistance systems have also become research hotspots in recent years.The identification of road traffic signs is an important part of the realization of driverless and assisted driving systems.Only when road traffic signs are accurately detected and identified can the vehicle’s control system be able to respond in real time or pass information to motorists.Therefore,a system that can respond in time and accurately recognize road signs is of great significance for the development of intelligent transportation systems.This paper researches the recognition of road traffic signs,and introduces neural network as a classifier,which significantly improves the recognition rate of road traffic signs and the calculation efficiency of the algorithm.The main work carried out in this article is:(1)As there is no large and public traffic sign data set in China,the Belgian traffic data set is used to design and verify related algorithms.Data enhancement of the data mainly includes: according to the influencing factors such as weather interference in the signs in the data set,combined with the actual situation in China,using the defogging algorithm to perform defogging on the data pictur es.In view of the problem of different image sizes,the size of the images is normalized to reduce the interference of external factors on the recognition of the signs.(2)Combining the rapid development and outstanding achievements of convolutional neural networks in recent years,this paper builds a traffic sign recognition network based on convolutional neural networks,uses deep convolutional networks to extract image features,and introduces visual attention The force mechanism realizes the enlargement of the detail area of the traffic sign through the attention network,thereby achieving a more accurate classification of the traffic sign.The final recognition accuracy reached 98.3%.(3)An optimization for the training process of traffic sign recognition deep neural network is proposed.A series of algorithms are used to optimize the training process of the neural network,including pre-training to accelerate the speed of network convergence,changing the traditional gradient descent method to random gradient descent with momentum,and setting the learning rate of the network to follow the network.Iterate and change.The cost of making the network lose only a small amount of accuracy is a huge savings in computing and time costs.