165 A Robust Method for Real-Time Detecting and Counting People

The paper presents a cost-effective approach to automatic people detection, segmentation, tracking and counting, which is designed for a system with an overhead mounted (zenithal) camera. First the background subtraction is done using a running average-like background model, then k-means clustering with some constrains is used to enable the segmentation of single person and get his location, before segmentation a people number estimation in the scene is used as a priori. Tracking of segmented people is solved using motion prediction method. People counting system have been applied to people surveillance and management areas. Experimental results suggest that the proposed method is able to achieve very good results in terms of counting accuracy and execution speed. (F-score above 95% over 3 video clips, at full frame rate 30 fps).