Research And Design On Traffic Sign Recognition System Algorithm

Traffic sign recognition system is an important system of car-environment interaction in the internet of vehicles,which is a major application in computer vision field.The study of the traffic sign recognition system algorithm can also contribute to the development of other related fields such as face recognition and pedestrian detection.In the practical application of detection and recognition,speed and accuracy are a key factor to limit its use.Therefore,this paper studies the traffic sign detection and recognition system,and proposes a new traffic sign detection and recognition algorithm.In this paper,the theory of traffic sign detection and recognition is discussed first.The theory of feature extraction and machine learning are described in detail.Then,the traffic sign recognition system is divided into two parts: detection and identification.The algorithm of both parts is designed,and the algorithm is experimented and analyzed.In the part of traffic sign detection,we uses the method of Contrast Limited Adaptive Histogram Equalization to enhance the image first.And then uses the Aggregate Channel Features to extract the features of the image,while the use of fast feature Pyramid algorithm to speed up the extraction of features and traffic signs image detection speed.In the post-processing part of the detection,the non-maximal suppression algorithm is used to fuse the detected windows,and the high accurate detection results are obtained.The experiment shows the effectiveness of the algorithm from the precision-recall curve.In the part of traffic sign recognition,a recognition method is proposed based on the fusion of the spatial pyramid descriptor and the histogram intersection kernel support vector machine on the identification of traffic signs.The appearance,color and contour information of traffic sign characteristics are described by extracting descriptor of gray pyramid histogram of visual words,color pyramid histogram of visual words and pyramid histogram of edge orientations gradients(PHOG)in this method.Spatial distribution of various features of the images can be well described by extracting the descriptor of the spatial pyramid histogram.Integrated them after extracting the spatial information of the image in terms of appearance,color,contour profile and feature,then strong robustness is embodied in space pyramid feature with integrated style.After the integrated feature is sent to HIK-SVM for training and classification,very high recognition rate could be achieved.