3D Human Pose Estimation from RGB Images

Abstract

Human pose estimation has received a lot of attention in the past years, also thanks to the advent of deep neural networks and their capability of detection patterns like human joints in images. Human Pose can be used for human-machine interfaces (HMI) but also to transfer motion from a real person onto a virtual avatar or character (Motion Capture). Another motivation comes from the field of autonomous driving, where the pose of a pedestrian can give us further information about its attention (field of view) and intention (Stay or go).

In this project we want to estimate the 3D pose of a human, defined by the 3D position of all his joints, directly from a single RGB image.

We want to develop a fast and lighweight approach, which is able to handle multiple persons per image and estimate absolute 3D positions of the human joints.

Requirements

The training of ML components, the evaluation and post processing are mostly done in Python.
Therefore basic Python skills are required.

Literature

2017-Coarse-to-Fine Volumetric Prediction for Single-Image 3D Human Pose
2019-XNect Real-time Multi-person 3D Human Pose Estimation with a Single RGB Camera

If you are interested , please contact us via e-mail:

Nikolas Brasch

Federico Tombari