Traffic Sign Detection And Recognition In Natural Scenes

In recent years,the rapidly expanding traffic network and rising car ownership have brought a series of traffic and environmental problems.In the face of the above problems,the intelligent transportation system has received more and more attention.As an important part of the intelligent transportation system,the Traffic Sign Recognition system has always been an important subject in the field of scientific research and business.The system has a very important role in areas such as assisted driving,unmanned vehicles,daily management of traffic management department,and multidisciplinary research.However,the actual environment is more complex,and the design process of the traffic sign recognition system faces some challenges,including: the weather and the lighting conditions are complex in natural scenes,vegetation and other obstacles are numerous.However,the traffic sign itself has a wide variety of shapes and types,and the existing sign recognition method with a high recognition rate has a complex structure and weak real-time performance.Therefore,this article focuses on the construction of a traffic sign recognition system in a natural scene.The specific research work is as follows:(1)In the image pre-processing and color segmentation of the recognition system,the issue of different weather conditions in the natural scene and the the impact of vegetation noise for segmentation are studied.Firstly,the weather conditions in the natural scene are judged by the luminance histogram,and the pre-processing operation based on the Retinex algorithm is adopted for the abnormal weather conditions.Then,on the basis of the normalized RGB threshold method,combined with vegetation threshold segmentation method and chromatic / achromatic decomposition method,a traffic sign segmentation algorithm in natural scenes is proposed.In the experimental part,we qualitatively and quantitatively compared and analyzed the enhancement effects of the three image preprocessing methods on the scene images containing signs.This paper compares the proposed segmentation method with common methods in the background interference,uneven illumination and other natural scene images.(2)In the traffic sign detection module of the recognition system,this paper proposes a sign detection algorithm based on the fusion feature,aiming at the detection problems of circular,triangular and octagonal shapes in natural scenes.In this dissertation,a Gist-LBP fusion feature describer is proposed after studying Gist,histograms of directional gradients,and local binary pattern feature.Experiments were carried out on the constructed sign sample database and the German traffic sign detection benchmark data set.It was proved that the proposed method is more universal and has higher accuracy than the single feature for detection of various shapes.(3)In the recognition module of traffic sign recognition system,in order to solve the problem that the existing sign recognition method has high recognition rate but complex structure,the paper proposes a traffic sign recognition method based on the local receptive field extreme learning machine optimized model.The parameters applied to the natural scene are searched using a grid search strategy,and then the affine transformation is used to solve the impact of unbalanced data on the model.According to the six categories of traffic signs,the recognition results of the traditional method and the proposed method are analyzed.According to the results of individual recognition,the changes of the recognition rate before and after the augmentation of the data are analyzed.The time comparison between the proposed method and the state-of-art method with high recognition rate is also presented.Multiple experiments show that this method has high recognition performance in the case of low time complexity.(4)In the system construction of sign recognition system,the application platform of traffic sign recognition system was constructed using WPF and MATALB.The application platform integrates multiple functional modules such as image pre-processing,segmentation,detection,and recognition.It realizes the functions of automatic segmentation,detection and recognition of multiple traffic signs in natural scenes,providing a basic experimental platform for research on traffic detection and recognition.