Research Of Pedestrian Detection Based Adaboost And Bayes Algorithm

With the rapid growth of quantities of vehicles and road accidents, how to protect the pedestrian effectively, has received considerable attention. An effective method is to detect the pedestrian advance. The research covers wide filed of pattern recognition, image processing, computer vision, probability and statistics. The existing pedestrian detection algorithms are generally affected by lighting, gesture, background and so on. In this paper, we do some research on pedestrian detection using Adaboost and Bayes algorithm.Adaboost is one of the most important algorithms of machine learning theory, and its basic idea is to form a strong classifier that has very strong classification ability using some general weak classifiers through a certain method. Bayes classifier is one of the most important statistical Machine Learning classifier, and its basic idea is to calculate the posterior probability based on the priori probability. This paper proposes a new classifier structure called coarse to fine based on Adaboost and Bayes algorithm. The main contributions of this paper are as following:First, according to the background of pedestrian, this paper divided the pedestrian into two modes:low contrast pedestrian, high contrast pedestrian. The high contrast pedestrian was divided into two modes (small angle pedestrian and big angle pedestrian) according to the angle of pedestrian legs separating to each other. For each of the modes, this paper trained a cascade Adaboost classifier. These three classifiers can reduce the complexity of the pedestrian, and reduce the difficulty of machine learning.Second, this paper trained the Bayes classifier using the confidence generated by Adaboost classifier. Each strong classifier in the Cascade Adaboost tests a sample independently, without sharing any information. The Bayes classifier can address these deficiencies using all the information generated by Adaboost classifier. This can reduce the waste of information and improve the recognition accuracy.Finally, this paper designed some method to improve the detection speed.