Multiple human tracking on robot operation system

Abstract

This paper proposes an approach based on robot operation system (ROS) to track group of human. Firstly, a method is proposed to fuse multiple human detected by Kinect cameras of Perception Sensor Network (PSN) system, which consists of Kinect cameras, and pan tilt zoom (PTZ) cameras. Then to track multiple human, grouping and ungrouping algorithms are proposed. When a group of human staying close together is formed, invariant features such as 3d human location, height, color and SURF feature points of human region of interest (ROI), which are identical criterions retrieved from both depth and color images of group members, then stored and updated into the group database. Based on the distance between a group location and each member location in the group, the system decides whether to keep this member in the group or to ungroup it from the group. If one ungroups from group the right name identification from the group database is reassigned for him by select the name yielding the least different of criterions. The experimental results demonstrate the outperforming of the proposed method in various scenarios.