Computer Vision & Graphics Group

Introduction

This dataset is focused on multiview captures of full-body humans in motion. Two subjects have been captured with two different hardware setups. On this website we provide the captured image data along with the camera calibration information as well as segmented per frame 3D reconstructions.

Acquisition and 3D reconstruction

For capturing the multiview sequences two different setups have been used (all hardware-synchronized):

• ATV: 8 Basler ace acA2000-50gc cameras, 25 fps, resolution 1920x1080, grouped to 4 pairs
• HBR: 12 Ximea CB200CG-CM cameras, 25 fps, resolution 5120x3840, grouped to 6 pairs

Both setups cover the frontal view onto the subject, ~145°. Each sequence is provided as sequence of raw images per camera and reconstructed 3D mesh per multiview frame. The 3D reconstructions have been calculated using [1]. They consist of vertices, RGB color per vertex and triangle topology.

Preview and Examples

Each subject has been captured with each acquisition setup, resulting in 4 sequences. The performed motions are movement of all limbs and turning around so that nearly the whole human surface is visible for some period of time.

	#frames	ø #vertices	ø #triangles	preview video	example
atvl	981	662k	1243k	multiview / meshed	multiview / mesh
atvp	1076	625k	1164k	multiview / meshed	multiview / mesh
hbrl	751	980k	1795k	multiview / meshed	multiview / mesh
hbrp	689	1256k	2327k	multiview / meshed	multiview / mesh

Requesting the data

Please send an e-mail to P. Fechteler <philipp.fechteler@hhi.fraunhofer.de> to request access to this database. All submitted papers or any publicly available text using this dataset must cite the following paper:

Related publication

[1] Waizenegger W., Feldmann I., Schreer O., Kauff P., Eisert P., Real-time 3D Body Reconstruction for Immersive TV, In Proc. ICIP’16 (Phoenix, Arizona, USA, September 2016).