다중 카메라 기반 휴머노이드 로봇 자세 추정

Abstract

Accurate robot localization is essential for a robot to move in a space. The location of an articulated robot can be defined by its body’s pose in the world coordinate system. For articulated robots, the movement of joints and obstacles often lead to occlusion challenges. The method of estimating keypoints using a voting algorithm based on a single camera sensor can robustly estimate the pose of a partially occluded object. However single camera system has limitations in detecting the entire space due to its limited measurement range. In this paper, we propose a method to estimate the pose of a humanoid robot’s body based on multiple camera systems to overcome these issues. Experimental results confirmed that the outputs of multiple camera sensors are effectively fused, and the pose can be robustly estimated even in situations where the robot is occluded by obstacles.