Master Thesis Proposal
“Robust Hand Pose Estimation with a RGBD Camera using Large Database"
 

I am looking for a Master student who is interested in writing his/her Master's thesis on related topics to “Robust Hand Pose Estimation with a RGBD Camera using Large Database".

 

Motion capture of human movements is prerequisite for realizing skill transfer from humans to robots. Although commercial motion capture systems are available for human whole body and hand motion capturing, these systems are expensive or require a special set up in the environment. Moreover, human subjects have to wear optical markers or additional sensors on their body. New technology for onboard camera systems is required to replace the distributed cameras commonly used for human motion capture [1][2]. In this thesis a method for three dimensional human hand pose estimation, capturing using a RGB-D camera like the Microsoft Kinect [4], should be developed [3].

 

Detailed research issues are
• Review of state of the art in markerless human hand motion capturing
• developing a new algorithm for 3D hand pose estimation using a RGBD camera
• (Offline) Learning with a large database
• Real-time computation of pose estimation
• Handling self-occlusion and connected fingers
• Test with multiple subjects’ hands
• Scale and Orientation Invariant pose estimation
• Evaluation of implemented algorithms in real experiments
• Comparison with an existing method [5]

Related works

[1] Azad, P., Ude, A., Asfour, T., Dillmann, R., Stereo-based Markerless Human Motion Capture for Humanoid Robot Systems, Proceedings of IEEE International Conference on Robotics and Automation 2007, p.3951-3956.

[2] Dongheui Lee and Yoshihiko Nakamura, Motion Capturing from Monocular Vision by Statistical Inference Based on Motion Database: Vector Field Approach, 2007 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS’07), 2007, p. 617-623.

[3] Jamie Shotton, Andrew Fitzgibbon, Mat Cook, Toby Sharp, Mark Finocchio, Richard Moore, Alex Kipman, and Andrew Blake, Real-Time Human Pose Recognition in Parts from Single Depth Images, CVPR, 2011

[4] Microsoft Corp. Redmond WA. Kinect for Xbox 360.

[5] Qixun Liang, Real-time Hand Pose Estimation with a RGB Depth Camera, master thesis, Technische Universit¨at M¨unchen, 2011 

 Requirements
- good programming skills on C and C++.

- basic knowledge of robotics and computer vision.

 
Thesis Supervision
Supervised by Prof. Lee at EI and Dr. Sturm at IN, TUM.
 

If you are interested in this topic, please contact Prof. Dongheui Lee (dhlee@tum.de). Additional information (e.g., academic transcript and state your skills) would be helpful.