Research of cues selection in pedestrian detector based on adaboost

Which cues to be used in describing pedestrian are the key to detect pedestrian with Adaboost algorithm. In this paper we presented 4 triangle cues and 16 complex cues by researching on pedestrianpsilas figures and proposed the way to calculate these cuespsila sum. To evaluate this calculating method, we proved the relation of iterative times with error and time spending. Compared with rectangle and asymmetry cues, triangle and complex cues are better in describing local characteristics of peoplepsilas appearance and designing pedestrian detector. The experiment indicated that triangle and complex cues can be used in constructing robust classifier to detect pedestrian.