Follow me: A simple approach for person identification and tracking

摘要

In this paper we present an approach to detect and keep track of a person walking in front of a mobile robot without any form of prior training. Additionally, the robot can be easily used by different persons. The only thing required to switch between them is a simple initialization step. The robot only uses a person's legs for the purpose of tracking. The approach is based on the fusion of sensor data from an ordinary RGB camera and a laser rangefinder (LRF). The image processing algorithm tries to identify the person's leg within the image stream, while the LRF is used to verify the position of the person with respect to the information provided by the camera. The system is implemented in ROS.