Research On Road Sign Recognition System Based On Deep Learning

In recent years,with the continuous improvement of people’s economic level and quality of life,higher requirements have been placed on the safety and comfort of automobile driving,which has also promoted the application of assisted driving technology in the automotive field.The vehicle assisted driving system can not only reduce the mental stress during the driver’s driving process,but also greatly reduce the probability of traffic accidents.The core element in the assisted driving system is the identification of road signs.With the development of science and advances in technology,deep learning techniques can improve the accuracy and efficiency of road sign recognition.The research in this paper is based on deep learning to explore the problem of road sign recognition,research and implementation of road sign recognition system.In order to implement the system,the work done in this paper is mainly:1.Based on the color characteristics of the road signs and the characteristics of the human eye’s color perception,the CLAHE algorithm is combined with the image enhancement algorithm based on Grey-Edge to effectively improve the image quality.2.Using the Leaky ReLU activation function,the residual network and R-FCN network improve the network structure of Faster R-CNN,use the latest Nadam optimization algorithm and OpenBLAS to speed up the network training speed,and improve the performance of the algorithm through experimental verification and analysis.Finally,from the four networks of experiments,the road marking detection network with high recognition accuracy and fast recognition efficiency was selected: R-FCN + ResNet-50.3.In order to improve the practicability of the R-FCN + ResNet-50 network,the RPN network is improved,and the number and size of candidate areas are reduced,which not only improves the accuracy of road sign recognition,but also improves the efficiency of recognition.4.Build a road sign recognition system,including software and hardware platforms.The improved R-FCN + ResNet-50 experiment on the offline video collected by the driving recorder was completed,which verified the reliability of the network model and the practicability of the system.