The Naked Truth:
Estimating Body Shape Under Clothing

Alexandru O. Balan and Michael J. Black
Brown University

Abstract

We propose a method to estimate the detailed 3D shape of a person from images of that person wearing clothing. The approach exploits a model of human body shapes that is learned from a database of over 2000 range scans. We show that the parameters of this shape model can be recovered independently of body pose. We further propose a generalization of the visual hull to account for the fact that observed silhouettes of clothed people do not provide a tight bound on the true 3D shape. With clothed subjects, different poses provide different constraints on the possible underlying 3D body shape. We consequently combine constraints across pose to more accurately estimate 3D body shape in the presence of occluding clothing. Finally we use the recovered 3D shape to estimate the gender of subjects and then employ gender specific body models to refine our shape estimates. Results on a novel database of thousands of images of clothed and “naked” subjects, as well as sequences from the HumanEva dataset, suggest the method may be accurate enough for biometric shape analysis in video.

Paper: [ PDF ]

Example

Our estimated shape model	Comparison with the visual hull	Visual hull from silhouettes

Clothing Dataset

6 subjects perform between 6-10 trials, each trial with a different set of clothing. There are a total of 53 trials and each trial consists of 11 predefined poses. Only half of the poses in each trial are shown in the videos.

Video: Single-pose Shape Fitting
Video: Batch Shape Fitting within each trial 

The faces of all subjects in the Clothing dataset have been hidden to preserve anonymity.

Videos are encoded using the Xvid codec. If you cannot see the videos in Windows, we recommend installing ffdshow.

HumanEva-II Dataset

HumanEva is a publicly available dataset of human motion capture data.

A kinematic tree tracker using a coarse cylindrical body model is used to track two sequences and provide initial pose estimates. Pose and shape estimates are refined for each frame independently of the other frames.

Video: Subject 2 - Single-pose Shape Fitting
Video: Subject 4 - Single-pose Shape Fitting

10 frames were selected from each sequence and shape was jointly optimized across poses.

Video: Subject 2 - Batch Shape Fitting
Video: Subject 4 - Batch Shape Fitting