Count passengers based on Haar-like feature in elevator application

Present elevator control use button sensors to determine when and where to dispatch an elevator car, which don't use the number of passengers. In this paper, we analyze images from camera to detect how many persons waiting for the elevator or in an elevator. A novel framework is proposed for optimized elevator schedule. Extended Haar-like features and Adaboost are used to train a head-shoulder classifier. Some images are selected from video according to elevator button callings to detect head-shoulder. To reduce false alarms a post process is added after detecting. Experimental results show, the proposed method with post process has higher performance than existed methods. The information of passenger number can he send to elevator control system for effective schedule, which can reduce passengers waiting time and elevator's unnecessary stop, finally save energy and reduce maintain fee.