Voting Based System for Robust 3D Hand Pose Estimation and Tracking

Abstract

Low cost depth cameras and advancements in the field of deep learning have paved the way to precisely estimate 3D hand pose using a single depth camera. However, to accurately estimate the pose one has to detect the hands in the scene and track them over consecutive frames. In this paper, we propose a voting based system to track and estimate the 3D pose of a human hand. Based upon Robot Operating System (ROS), it comprises a hand segmentation stage, a clustering stage, a voting stage, a validation stage and a pose estimation stage. The final output is the 3D pose which is then used by a robot to follow the human hand.