Extraction of 3D Pose in Video for Building Virtual Learning Avatars

摘要

From an image of a person, we can easily guess the 3D coordinates of the body parts. This is because we have acquired a 3D mental model from observing humans and interacting with them. This capacity easily achievable for humans is not systematic when it comes to computers. In this paper, we describe an approach that aims at estimating poses from video with the objective of reproducing the observed movements by a virtual avatar. We propose the fragmentation of submitted videos into series of RGB frames to process individually. We aim two main objectives in our work. First, we achieve the extraction of initial 2D joints coordinates using a method that predicts joint locations by part affinities (PAFs). Then we infer 3D joints coordinates based on a human full 3D mesh reconstruction approach supplemented by the previously estimated 2D coordinates. Secondly, we explore the reconstruction of a virtual avatar using the extracted 3D coordinates with the prospect to transfer human movements towards the animated avatar. This would allow to extract the behavioral dynamics of a human, allowing to detect some health problems, for instance in Alzheimer. Our approach consists of multiple subsequent stages that show better results in the estimation and extraction than similar solution due to this supplement of 2D coordinates. With the final extracted coordinates, we apply a transfer of the positions (per frame) to the skeleton of a virtual avatar in order to reproduce the movements extracted from the video.