Traffic Sign Recognition Research Based On Convolutional Neural Network

With the development of intelligent driving,the detection and recognition of traffic signs have become one of the important functions of vehicle intelligent technology.The detection effect of traffic signs is affected by the complexity of real traffic scenarios such as lighting conditions and occlusion,as well as the high accuracy and real-time requirements of vehicle detection in driving,it brings great difficulty to the research of traffic sign detection technology.Traditional traffic sign detection algorithms mainly adopt artificial design features to obtain target candidate areas.it is difficult to realize real-time road driving detection because of the large amount of computational data in the detection process.Target detection based on Convolutional Neural Network(CNN)can automatically extract features and has a small calculation burden and the function of detecting multiple types of targets,which brings new technologies to traffic sign detection.This paper proposes an improved deep CNN model algorithm for Chinese traffic sign detection.The main research contents are as follows:First of all,establish a Traffic Sign Data Set in line with the actual situation of China’s traffic roads in view of the lack of China traffic sign data set.It adopts area cropping,histogram equalization,size normalization and image contrast enhancement to preprocess the image,and it also adds BM3D(Block Matching 3D)denoising algorithm for denoising,which reduces the effect of the real environment on image quality and the high-quality images are obtained.Furthermore,the effectiveness and correctness of the preprocessing method are verified through experiments.Secondly,significantly improve the detection accuracy based on YOLOv2(You Only Look Once v2)algorithm and CNN.Due to its low detection accuracy for small targets,it adopts normalization to the loss function of the original network and gives a weight value according to the loss of each part to increase the focus on small targets;then it discards 4 layers of 3×3 convolution in YOLOv2 network structure,and adds a layer of 1×1 convolution,which reduces the network parameters and the calculation time,and improves the detection speed of the algorithm;it also introduces multi-scale feature fusion to fuse high-level features into low-level feature maps to improve the recall rate for small target detection.Finally,build an intelligent traffic sign detection system and complete the detection experiment.It adopts Tensorflow as the framework.According to the improved detection algorithm,it conducts detection experiments through the self-made N-CTSD(New-Chinese Traffic Sign Detection)China traffic sign data set.The test results show that the improved model has mAP and FPS of 84%and 52 FPS on the N-CTSD dataset,respectively,which are improved by 5%and 6 FPS compared to the original model.And it tests on GTSDB(German Traffic Sign Detection Benchmark)to verify the robustness of the model.The mAP and FPS are 83%and 51FPS,which are increased by 7%and 6 FPS,respectively.The research of this paper is of great significance to improve the accuracy and real-time performance of traffic sign detection,and plays an important role in promoting the development of intelligent driving technology.