Robust people counting using a region-based approach for a monocular vision system

Abstract

This paper presents an automatic people counting system for indoor environment. The system processes visual information obtained from an IP camera that is mounted in topview position above the door. The people detection process is designed to perform as a sequence of steps, including background subtraction, morphological operation, and blob detection. The results from the detection process are filtered and tracked to improve robustness of the system. A region-based approach is applied to the people counting process. The image obtained from the camera is divided into three regions and the counting algorithm is developed to track the people's state in the regions. The algorithm can count people with IN and OUT direction. Even when the system was setup without lighting control, the developed system can correctly count people as shown in the experiments. The output of the system can be exported to be used in downstream applications such as security surveillance and statistical analysis of customer behavior.