Abstract:

In view of  the problem of the low degree of single-eature recognizability in pedestrian detection based on feature, this paper introduces a kind of algorithm，AdaBoost for pedestrian detection based on multi\|feature, and presents a new integrated approach of multi\|feature which is the fusion of gray scale and contour information. This approach establishes a classification model by the histogram statistics of the weight samples, and the probability distribution is represented by multiplication of several histograms. So  the joint probability based on multi\|feature can get more accurate description of pedestrian, and improve the robustness of pedestrian detection. The experimental results show that pedestrian detection in this paper has improved the detection rate, lowered false alarm rate in a large extent, and the confidence level of target recognition has been markedly improved, and it can get the better performance of pedestrian detection in various natural background.

Key words: pedestrian detection, multifeature, histogram stat