Real-Tim 3D Reconstruction/Fitting of Human Voxel Models

German Cheung, Takeo Kanade

Please contact German Cheung at german@ux2.sp.cs.cmu.edu for further details

Return to Homepage

RealTime 3D Reconstruction

Temporal SFS

Human Kinematic Modeling and Motion Capture

Human Motion Transfer

Objective

This project analyzes the effect of silhouette noise on the accuracy of the voxel-based Shape-From-Silhouette algorithms. Based on the analysis, a fast algorithm called SPOT for voxel projection/testing is derived. Using SPOT, a real-time system is built to reconstruct 3D voxel models of humans. Simple ellipsoidal shells representing different body parts are also used to fit the reconstructed voxels. Technical details of this project can be found in the documents listed below:

CVPR2000 paper titled `` A Real Time System for Robust 3D Voxel Reconstruction of Human Motions''
(Download: pdf)
Chapter 3 of German Cheung's PhD thesis titled ``Visual Hull Construction, Alignment and Refinement for Human Kinematic Modeling, Motion Capture and Rendering''
(Download: pdf (7.94M))

Real-time 3D Human Voxels Reconstruction/Fitting System Using SPOT

The SPOT Algorithm	System Architecture	Screen Shot

Results

Video One: The System in Action (Download: mpeg)	Video Two: Screen Captured Results (Download: mpeg)

Real-Tim 3D Reconstruction/Fitting of Human Voxel Models

German Cheung, Takeo Kanade

Objective

Real-time 3D Human Voxels Reconstruction/Fitting System Using SPOT

The SPOT Algorithm

System Architecture

Screen Shot

Results

Video One: The System in Action (Download: mpeg)

Video Two: Screen Captured Results (Download: mpeg)