Interactive Indoor Environment Mapping Through Visual Tracking Of Human Skeleton

This paper presents a novel human-robot interaction approach to grid mapping of an indoor environment based on a 3D kinect sensor and the grid-based mapping algorithm. It mainly includes three modules: skeleton tracking, robot control and GMapping. Firstly, the skeleton tracking module builds a human skeleton model, extracts the skeleton joints' position information from 3D visual data and generates digital signals through identifying some simple motions and events. Then according to different digital signals and joints' position information, the robot control module enables the robot to take different actions such as following the person, stop and so on. Finally, the grid map of the environment is built through GMapping algorithm based on odometry and laser data, which is improved by Rao-Blackwellised particle filters. The proposed approach has been implemented successfully in several different buildings and can be applied to service robots. Compared with traditional roaming for mapping, human guiding the robot for mapping is more efficient and takes less time in a complicated environment. Meanwhile, compared with wearable motion sensors attached to the human body, this approach is more convenient and make the user more comfortable.