Predicting Reaching Targets using Human Skeletal Pose Estimation

Abstract

The paper describes a method for prediction of reaching targets by measuring 3D human pose with RGB-D camera. The method uses an RGB-D camera to combine 2D human pose estimation on the image with depth information to estimate 3D human skeletal pose. A future pose is predicted by learning human motion using RNN. By using the future pose prediction and decision tree, the method predicts which object on the desk is a reaching target. The experiments are performed for motion of reaching targets on a desk.