Research Of Fast Pedestrian Detection Algorithm In Intelligent Transportation System

With the rapid development of modern society, the fast urbanization brings traffic tremendous pressure. in order to reduce the incidence of traffic accidents, keep the traffic flowing continued effective operation, it needs effective management. Traditional manual way spends a lot of financial and material resources. And it has low efficiency. intelligent transportation systems has become an inevitable choice for the future transport development. In intelligent transportation systems, pedestrian detection has important practical significance, intelligent detection technology can be applied to vehicles, reduce traffic accidents. Applied to the monitoring system, you can analyze the behavior of pedestrians.Pedestrian detection algorithm emerges in endlessly, it can be divided into two categories: one is based on the traditional template matching algorithm. The other is the image feature extraction detection algorithm. This paper studies the image feature extraction algorithm. Through the study of pedestrian detection algorithm, we can effectively improve the detection rate in detection system and reduce the detection time. This paper studies the existing pedestrian detection algorithm, and has a detailed analysis of existing algorithms, makes a in-depth comparison between the advantages and disadvantages of various algorithms. Existing algorithms can be divided into three categories: the global-based detection, the multi-site detection and multi-angle detection. The research focus of this paper is in front of the two categories. In order to address the diversity of the physical characteristics of pedestrians, pedestrian movement arbitrary rule, select a single feature of the most powerful features of HOG, the HOG gradient histogram with SVM support vector machine combined as the pedestrian detection algorithm, the experimental results show that the HOG + SVM algorithm has good recognition rate. When classifier training, the traditional HOG feature extraction has higher dimension, there is a lot of redundant information, so it has more complex algorithm. To overcome this shortcoming, the PCA main analysis is introduced into the improved Algorithm, and format a new PCA-HOG features. After the PCA algorithm can effectively reduce the processing feature of HOG overlap of information, and reduce the dimension of the feature space. By comparing the experimental results of single HOG features, classifier training and testing time of PCA algorithm processing are greatly reduced, and the detection accuracy has also been improved.Previous analysis HOG + SVM algorithm is based on the overall global detection algorithm, a pedestrian under shelter, we have excellent detection results. In the detection process, the detection window scans the information of the image one by one, extracts all eigenvalues to determine whether there are pedestrians. this overall detection algorithm in the case of occlusion detection effect is not ideal. In order to improve the detection rate and detection time in the case of pedestrian shelter. we proposes multi-feature fusion of HOG and LBP. This feature both has the pedestrians edge gradient information, but also has the texture feature information, it can effectively compensate for deficiencies single HOG features place. Since HOG feature extraction calculation is slow, a fast detection algorithm is proposed, integral image of HOG feature extraction methods can improve the detection speed. In the classifier designing, we design two classifiers, a general classification and a bust classifier, when detect the pedestrian.we make the image feeding into the general classification to get a judgment, if the judgment is no, then feeding into the bust classifier to judge whether pedestrian exist. The experimental results show that under shelter, the algorithm has excellent detection results. The previous theoretical analysis, we propose PCA dimension reduction of HOG features, while incorporating LBP feature. comparing the experimental results, the algorithm can effectively improve the recognition rate, and also improve the training and testing speed.