Design and implementation of a human tracking and following capability for a robotic avatar

• Main Author: Yao, Xiling.
• Other Authors: Seet Gim Lee, Gerald
• Format: Final Year Project
• Language: English
• Published: 2013
• Subjects: DRNTU::Engineering::Mechanical engineering::Robots
• Online Access: http://hdl.handle.net/10356/54019
• Institution: Nanyang Technological University

In this project, the human-following capability has been implemented on a mobile platform (MAVEN-II) which was also called a robotic avatar. The design and implementation of human-following capability was part of the research project under the general topic of Telepresence, which aimed at developing remotely controllable robots that can interact with humans in social scenarios (conference, eldercare etc.). Previous designs had used RGB cameras and laser range finders to detect and track humans, with Mean Shift, Histogram of Oriented Gradient (HOG), and Background-Foreground Model as the commonly used algorithms. However, these methods were not able to re-identify the targeted person among a crowd when he/she came back to the scene after being occluded or being out of the sensor’s field of view. In this project, the Kinect sensor was used to detect the person and to track his/her position in 3D space. In each frame of the sensor reading, the histogram in HSV color space was calculated for the targeted person’s clothes, while the person’s position was updated and stored until he/she was in a situation (being occluded or away from the field of view) where he/she could not be detected by the Kinect sensor. From then on, for every person detectable by the sensor, his/her current color histogram and 3D position was compared with that of the originally targeted person with a Matching Level calculated. Hence if the targeted person re-enter the field of view, he/she would generate the best Matching Level and then he/she would be recognized as the target. Based on the relative position from the robot to the targeted person, the robot’s linear and rotational travelling speed was calculated, and it was able to follow the person when he/she was walking around. During the human-following process, the robot tried to maintain a constant distance to the targeted person. In addition to the mode of human-following at the person’s back, which had been implemented in previous research, the Side-by-side Following mode was implemented in this project, which enabled the robot to accompany the targeted person by his/her side when he/she was walking around. Obstacle avoidance function was also implemented using the Virtual Potential Field concept. Experiment result showed that the Back-Following function was satisfactorily robust; while the robot movement in the Side-by-side Following mode was unstable.