An Improved Pedestrian Detection Algorithm Based on AdaBoost Cascading Stucture

Pedestrian detection is a very difficulty in Intelligence Traffic System (ITS). For more rapid detecting pedestrian, given that Adaboost algorithm has features of simpleness, reliability and high learning accuracy, and this paper proposed a real-time detection approach based on Adaboost. Firstly, adopt section-distribute model to detect the region with pedestrian in the frame of the video, then analyze the candidate regions, and determine whether it is pedestrian or not. Therefore, we need use more features to implement pedestrian modeling. This paper took the characteristics of rectangular edge description method as a reference to analyze the characteristics of pedestrian attitudes and gained new features - triangle characteristics. Through training with Adaboost algorithm to get an ideal and high accuracy recognition pedestrian classifier. Upon this basis, this paper improved the strategy of sample weight adjustment, reducing the phenomenon of overfitting. The experiment result shows, the approach can help to rapidly and precisely detect pedestrian online, and be more real-time.