Study On The Method Of Pedestrian Detection In Automobile Safety System

Because of the rapid growth of quantities of vehicles and road accidents, how to reduce road accidents has become the focus of attention. As a technology of vehicle active safety, pedestrian detection system has been concerned and studied by more and more departments and institutions. Pedestrian detection is the combination of the technology of sensors, image processing, computer, and so on. It can detect whether there are pedestrians in the aim area, and reduce road accidents between pedestrians and vehicles.This paper has studied the method of pedestrian detection in front of the vehicles based on statistical learning. Pedestrian detection based on statistical learning consists of two parts of the sample feature extraction and classifier selection. This thesis describes the pedestrian features and classifiers used in the paper firstly, which have an improvement. The Haar-like feature types are increased to improve pedestrian description ability, and a new method is introduced to improve the training speed of the weight update in the Adaboost algorithm. Then in order to make the pedestrian detection system fast and accurate, we get two graded pedestrian classifiers by combing with pedestrian contour features and integrating the advantages of Cascade which has a high detection speed and SVM which has a better robustness.This paper has studied a graded pedestrian detection algorithm based on Haar-like features and Hog features. Firstly, ten types of Haar-like features are given, and they are trained through Cascade, we can get the coarse classifier and determine the pedestrian candidate area in front of the vehicles. Then, we select the Hog features of strong ability to distinguish pedestrians by using SVM of different kernels and get the fine classifier to define pedestrian orientation.Besides, this paper has studied a graded pedestrian detection algorithm based on Hog features, texture features and invariant moment features. Firstly, we determine the pedestrian candidate area by combing Cascade with Hog features. Then, we select texture features and invariant moment features as pedestrian features and get the classifier to identify pedestrians by using SVM of different kernels.Test results show that the proposed methods can identify the pedestrians of different sizes, colors and shapes who are in front of the vehicles effectively, they have a higher detection rate and achieve a better detection effect.