Robust Articulated-ICP for Real-Time Hand Tracking

• YouTube Video: https://youtu.be/rm3YnClSmIQ
• Paper PDF: http://infoscience.epfl.ch/record/206951/files/htrack.pdf

Disclaimer

To obtain the results shown in the video (or as seen in the live demo sessions at SGP'15) proper hardware is necessary:

• Ubuntu 14.04 (on windows/osx the OS interferes with GPU on Compute/Graphics context switches)
• Intel Core i7 @4GhZ
• CUDA Graphic card (NVIDIA GTX980 used in our demo)
• Senz3D depth camera (SENSOR_DEPTHSENSEGRABBER, use SoftKinetic or Intel/Creative)

Other notes:

• note the software must be compiled in 64bits!
• Wristband color calibration (make sure the wristband is detected robustly otherwise the tracking might not perform as effectively, you can check this by enabling "show wband" in the htrack_atb application)

BibTex

@article{htrack_sgp15,
author = {Andrea Tagliasacchi and Matthias Schr{\"o}der and Anastasia Tkach and Sofien Bouaziz and Mario Botsch and Mark Pauly},
title = {Robust Articulated-ICP for Real-Time Hand Tracking},
journal = {Symposium on Geometry Processing (Computer Graphics Forum)},
month = July,
year = 2015}

News

• July 8th 2015: we won the Best Paper Award at the Symposium on Geometry Processing (SGP) 2015!
• June 12th 2015: official code release (in sync with CVPR'15 HANDS workshop)

Acknowledgements

• Thanks to "Tu-Hoa Pham" for his pull request that improved the SoftKinetic (Senz3D) Sensor!