Laser/vision-based tracking of people with sudden changes in motion

Abstract

This study presents a laser/vision-based tracking of people in 3D space. A sensor node consisting of a two-layered laser-range scanner (LRS) and fisheye camera is placed in an environment. LRS detects the heads and knees of people, and the camera captures the color images of people detected by the LRS. From the position data of people scanned by LRS, peoples' motion in the horizontal plane is tracked, and head motion in the vertical direction is tracked on the basis of color images obtained by the camera. People are tracked via a model-based tracker; the interacting-multiple-model estimator and multiple-model particle filter are applied to track people whose motion suddenly changes, such as walking, running, turning, jumping, and crouching. Experimental results of the tracking of three people moving randomly in 3D space validate the effectiveness of our tracking method.