Traffic Signs Recognition Based On Deep Learning

Traffic signs are important sources of information obtained during the driving of vehicles.Accurate and rapid recognition of traffic signs is of great significance for ensuring traffic safety,traffic order,and improving traffic efficiency.It is also of great significance to the currently emerging advanced driver assistance system and unmanned research.Therefore,traffic sign recognition is an important research direction of the Intelligent Transportation System(ITS).Under natural road traffic scenarios,weather conditions,light intensity,viewing angles,or occlusion have caused many difficulties in the detection and identification of traffic signs.Therefore,it is necessary to study the traffic sign recognition algorithm with high accuracy,robustness and good real-time performance.Recently,deep learning has been widely used in image and video processing,voice and audio processing,and has made rapid progress.This paper improves the detection and recognition algorithm of traffic signs based on deep learning and validates the effectiveness of the algorithm through a large number of experiments.The main research content is as follows:1.For the problem of traffic signs recognition rate is not high,this paper proposes Double Channels Layer-Skipping Convolutional Neural Network(DCLS-CNN)structure with different scales.The global features of the traffic signs are acquired on the channels one,and local features of the traffic sign are acquired on the channels two.The low-layer features and high-layer abstract features of traffic signs,which these features fusion in the full connected layer.Finally,the fusion feature input a classifier for traffic sign recognition.Trained and tested by the German Traffic Sign Recognition Benchmark(GTSRB),the influence of the setting of deep learning network parameters on the experimental results was analyzed,and the recognition rate of the algorithm reached 97.96%,which is significantly better than that of Layer-Skipping Convolutional Neural Network(LS-CNN)and hand-crafted methods.2.For the problem of traffic sign detection and recognition precision is not high,an improved SSD detection algorithm is proposed.Firstly,the high-level smaller-scale feature layers used for detection on the basis of the SSD algorithm are removed.Secondly,adjust the low-level feature map scale and width-height ratio to distribute more default boxes at lower levels,thereby enriching the fine features in the traffic sign scene map.Finally,collection of a large number of real traffic signs scene map as an experimental data set.The experimental results show that the proposed algorithm has good robustness to traffic with different environmental conditions,and the mean average precision(m AP)is 0.7528,and has raised about 10% compared to the classical SSD,which verifies the effectiveness of the proposed algorithm.