Application Of Traffic Sign Recognition Method In Driverless Cars

Unmanned driving technology involves a wide range of disciplines.When it has a technical breakthrough,it can promote the development of other disciplines.Therefore,the development of unmanned vehicles is more and more valued by society.The detection and recognition technology of the traffic signs contained in it is the guarantee for the safe driving of unmanned vehicles.Due to the influence of the external environment,such as lighting and motion blur,it has brought many problems and challenges to the actual detection.In order to meet the requirements of unmanned vehicles for traffic sign recognition,this paper uses image detection algorithm to detect the shape and color features of traffic signs,and then uses the convolutional neural network to classify and recognize the detected pictures.In order to meet the requirements of recognition accuracy and rate,this paper modifies the VGG-16 convolutional neural network,and proposes the VGG-8 convolutional neural network,which is simplified and optimized to increase its practicality.The accuracy of constructing the network can reach 96.54%,and the detection effect is better than before.Since the final purpose of this article is to complete the detection and recognition process on a moving car,it is necessary to have a faster detection speed while ensuring accuracy.In order to meet the requirements,this paper draws on the idea of SSD target detection algorithm and improves it.First,the backbone network structure of the SSD target detection framework is improved,the convolutional layer is deleted,and the network dimension is changed to the original one-half.By reducing the network structure,the detection speed can be accelerated to meet the real-time requirements.Requirements.Since the original network is not very effective in detecting small targets,the feature fusion module is modified to increase its ability to detect small targets.In order to verify the practicability of the algorithm in this paper,at a speed of 40 km / h,a high-speed motion camera is used to take real photos of the roadside road conditions,and the traffic sign pictures in the video are extracted as test samples.It has been verified that the algorithm described in this article has been improved from before.Among them,the accuracy of detection combined with image detection algorithm and VGG-8 convolutional neural network is higher,and the detection rate of SSD target detection framework is faster.