Road pedestrian detection based on a cascade of feature classifiers

How to detect pedestrian faster and more accurately based on video is the key to pedestrian detection. A method of pedestrian detection based on a cascade of feature classifiers is proposed in this paper. First, according to the different features between pedestrians and non-pedestrians, several special features are selected. Second, according to AdaBoost classifier training theory, several weak classifiers are trained using feature values extracted in sample space. Then the cascade sequence of weak classifier is determined by the rule presented in this paper. The final cascaded classifier is the combination of weak classifiers in a specific order. Experimental results illustrate that the cascaded classifier is effective for lowing false positive rate and ensuring high detection rate. Besides, a real-time detection is guaranteed by the high detection speed.