Convolutional Neural Network Powered Identification of the Location and Orientation of Human Body via Human Form Point Cloud

This paper proposes a Convolutional Neural Network (CNN) based scheme using the point cloud of human body to identify its location and posture. The point cloud is randomly generated but confined within a human form. The CNN-based model is fed with point cloud for predicting mass center location and orientation of the body with help of high end graphical processing units. We propose to project the point cloud in two vertical planes to exploit the image recognition capability of CNN. The proposed method is tested with a single person for three primary postures: standing, sitting and lying, to evaluate the prediction capability. Effects of the number of points indicating point cloud density and the distance between the observation station and the target are investigated. Simulation results show a body part dependent localization accuracy smaller than 8 cm, and posture dependent success rate above 93%, validating the functionality of proposed scheme.