People Counting across Multiple Cameras for Intelligent Video Surveillance

Abstract

Pedestrian counting is widely used in civilian surveillance. In this paper, we present a people counting system which estimates the number of people across multiple cameras with partial overlapping Fields Of Views (FOVs). The main contributions of this paper include: 1) we propose a multi-object detection and tracking method by means of synthesizing the local-feature-level information into object-level based on an electing and weighting mechanism (EWM), 2) We present a scheme to integrate the counting results from multiple cameras. Through homograpy transform and similarity measurement rules, the system can find the objects in overlapping FOVs and finally estimate the integrated number of people across multiple cameras. Experiments results demonstrate that our system is effective and accurate for multi-camera people counting.