A head position estimation method for a variety of recumbent positions for a care robot

Abstract

This paper describes a head position estimation method for a variety of recumbent positions for a care robot. First, the robot detects a fallen person using a laser range finder and moves there for getting a point cloud. The obtained point cloud is matched with those with head positions in the database using the ICP (iterative closest point) algorithm for estimating the current head position. Then, extracted head positions with matching scores above a threshold are clustered, and the center of the cluster with the highest accumulated scores is determined to be the head position. A human pose database construction system is also described.