Research On Traffic Sign Recognition Method Based On Multi-feature Fusion Of Block Weights

With the development of science technology and the improving of people’s living standards,intelligent transportation systems came into being which provide solutions for people to travel safely.Traffic sign is the most important of the intelligent transportation systems.It includes preprocessing,traffic sign detection and recognition.Traffic sign recognition has always been a hotspot for researchers and scholars.Although many years of research has achieved certain results,with the socio-economic development,the increase in car ownership will inevitably lead to traffic congestion.This paper mainly discusses and studies the three parts of traffic sign preprocessing,detection and recognition.At the same time,the feature extraction is improved for recognition.A method of information entropy weighted block LBP features is proposed.This method first calculates the block information entropy And the global information entropy,and then the ratio between the two to obtain the weight of each block,and then cascade the HOG feature,and then linear fusion to obtain the final feature,the effectiveness of this method is verified through experiments.This paper’s main work is as follows:(1)Pretreatment.Traffic sign preprocessing is aimed at the situation of insufficient light or rain and fog.This article will compare and analyze the histogram equalization and contrast limited histogram equalization of traffic sign images collected in insufficient light or fog.Experimental comparison under conditions,the use of contrast-limited histogram equalization,which enhances the image,is better than histogram equalization,so the finaluse of contrast-limited histogram equalization method for image enhancement.(2)Traffic signs detection.For the speed limit sign,the red region is extracted under the HSV color space,and the region of interest is segmented finally,and then through a binarization and morphology operation,a part of the ROI region is first filtered by the big area method,and finally based on the roundness and area The weight method accurately locates speed limit traffic signs.(3)Traffic sign recognition.This paper proposes an information entropy weighted block LBP feature method.Since one feature cannot fully characterize the traffic sign,this paper extracts multiple features and uses the complementarity between the features to more fully characterize the traffic sign information.First extract the gradient histogram HOG feature from the image,then extract the LBP feature from the image block,and use the information entropy weighting(the ratio of block information entropy to global information entropy)to calculate the weight for each block to obtain the BWLBP feature It is cascaded with HOG to form the fusion feature HOG-BWLBP.Experiments show that the BWLBP feature extracted after the block information entropy is weighted has stronger expressive power and better classification effect than the original LBP feature.Finally choose linear support vector machine for classification and recognition.