Review of 3D Sensing Modalities for 3D Pose Estimation

Three-dimensional (3D) pose estimation is an essential process in all human interactive systems. The capturing of 3D sensing modalities consists of time-of-flight sensors, structured light sensors and stereo sensors. These data can subsequently be processed for 3D human pose using deep learning approaches. These approaches can be classified into direct 3D pose estimation and 2D-to-3D pose estimation. The 3D sensing modalities and the respectively processed algorithms are highlighted to show the advantages and disadvantages. The production of the 3D human pose can be directly processed using the volumetric-based method, point cloud-based method and graph theory-based method. These methods are recently drawn huge attention in the field to increase depth processing and fast computation on pose estimation in 3D space.