3D Human Pose Estimation Using HG-RCNN and Weak-Perspective Projection
This project focuses on multi-person 3D human pose estimation from monocular images using advanced techniques like HG-RCNN for 2D heatmaps estimation and a shallow 3D pose module for lifting keypoints to 3D space. The approach leverages weak-perspective projection assumptions for global pose approximation, allowing for efficient estimation of body joint positions in Euclidean space. The proposal aims for sub-linear time complexity with in-the-wild performance, without requiring a specific multi-person 3D pose dataset.
Download Presentation
Please find below an Image/Link to download the presentation.
The content on the website is provided AS IS for your information and personal use only. It may not be sold, licensed, or shared on other websites without obtaining consent from the author. Download presentation by click this link. If you encounter any issues during the download, it is possible that the publisher has removed the file from their server.
E N D
Presentation Transcript
Multi-Person 3D Human Pose Estimation from Monocular Images Rishabh Rishabh Dabral IIT Bombay IIT Bombay Dabral Rahul Rahul Mitra IIT Bombay IIT Bombay Mitra Nitesh Nitesh Bharadwaj Bharadwaj IIT Bombay IIT Bombay Abhishek Sharma Abhishek Sharma Axogyan Axogyan AI AI Ganesh Ganesh Ramakrishnan Ramakrishnan IIT Bombay IIT Bombay Arjun Jain Arjun Jain Axogyan Axogyan AI AI
Problem Statement 3D Human Pose Estimation: Given an image of a human, estimate the 3D positions of body joints in Euclidean space.
Problem Statement 2D Human Pose Estimation: Given an image of a human, estimate the positions of body joints in pixel coordinates.
Problem Statement A Multi-Person 3D Pose Estimation: Sub-linear time complexity w.r.t. the number of persons In-the-wild performance
Proposal: HG-RCNN Estimate the 2D heatmaps using an Hourglass-augmented Faster-RCNN Lift the keypoints to 3D using a shallow 3D pose module. Approximate the camera-relative positions of the 3D skeletons Modular; does not require a multi-person 3D pose dataset. Datasets: MSCOCO (multi-person 2D) and Human3.6M (single-person 3D)
Global Pose Approximation Weak-Perspective Projection Assumption Assume that the 2D pose is the scaled version of the orthographic projection of the global 3D pose: ?2?= ??3? How to find ?? Ratio of Sum-of-bone-lengths as the scale factor ? : ? = ??3? ?2?, Y= ??3? ?2?, X= ??3? ?2?
