Based On Block Kernel Function Feature Of Traffic Signs Recognition

As part of the intelligent traffic system, the traffic sign recognition has a broad prospect of application. Meanwhile, as the typical issue of target recognition in natural scenes, the traffic sign recognition also has the academic significance. The complex of background, the diversity of traffic signs, and the factors of light, pollution, distortion, shelter and the real-time ability, all of which make the development of a new practically traffic sign recognition system very hard. This article proposes a method of traffic sign recognition, which is based on the feature of block kernel function, researches and improves the frequently used block method.This paper lodges the traffic sign recognition, which is based on the shape of the pre-classification process. Traffic signs involved in the experiments mainly are comprised of three kinds shapes, circles, triangles, and prismatic. Firstly traffic sign images is pre-processed, and then it is pre-classified based on the shape characteristics, finally the traffic sign is categorized in the samples of the same shapes. The Hough transform is used in the pre-classification of traffic signs, meanwhile, according to the characteristics of the traffic sign images, the processes of the shape discrimination and the selection of the parameters are analyzed and tested.The feature extraction is an important stage of traffic sign recognition. This paper focuses on the research of feature extraction method of block kernel function. According to traffic sign recognition, this paper proposes two methods which include the linear weighted method and the sub-mode combination method. The experiments shows that, in the case of the same number of training samples, the feature based on combination of sub-mode performs better than KFDA feature and the feature based on linear weighted of the classification results. Meanwhile, partitioning strategies in the two sub-block methods are analyzed and improved. In this paper, the linear weighted block kernel function analysis strategies of adjacent block and slide block. And the block kernel function of sub-pattern analyses strategies of adjacent block, overlapping the edge of the block and slide block. The classification results confirmed on the Nearest Neighbor Classifier shows the effectiveness of our method.The HOG feature describes the edge feature of image, and feature of block kernel function describes the statistical characteristics of the image. This paper researches the method of combining these two features to recognize the traffic signs. The combined method similar to voting is proposed, which needs to construct a posterior probability matrix to discriminate the category of traffic signs. This paper also analyses and compares the recognizing effect of extracting the feature of HOG and the feature of block kernel function by using k Nearest Neighbor Classifier and SVM classifier. The result of classifying43kinds of traffic signs show that the feature of block kernel function as well as the improved partitioning strategies in the multi-classifiers combined k Nearest Neighbor Classifier and SVM classifier achieve a good classification performance.