Development of the human tracking system using the cooperating of multiple cameras arranged sparsely

Abstract

In recent years, security cameras have been established in public facilities such as schools and shopping malls and various places such as public roads and general households. The data taken by the security camera is very useful because it is used in various situations such as tracking people and objects, verifying accidents and incidents, and observing the weather. The number of security cameras is more effective as the number increases, however, the landscape deteriorates and causes great stress to people. In this study, we propose a system that shares camera information in multiple sparsely arranged camera environments. In the proposed system, it is most important to select the features for each camera because the environment is completely different for each camera (e.g. light conditions, occlusion, camera direction). In this paper, we apply the proposed system to people tracking and show the system's usefulness by verifying various features. In human tracking, to solve the partial occlusion of the target, a particle filter is used. As the features for human tracking, the STHOG feature is adopted. The STHOG is a feature quantity that focuses on the walking characteristics of humans and is robust to fluctuations in lighting. We performed the simulation of the human detection, in the case where there are multiple people whose colors are similar to the target. As a result, The effectiveness of discrimination by STHOG features is shown compared with the case of only color information.