Extraction of low dimensional manifolds of human hand movements

The goal of the project is to extract low dimensional manifolds of human hand movements. This means that a low dimensional subspace is extracted from the 22 dimensional angular joint space of a human hand, where each dimension represents one joint of the hand. This is done by using state-of-the-art dimensionality reduction techniques.

The main task is to implement a specific dimensionality reduction technique in Matlab and extract the low dimensional subspace from a pre-recorded data set of human hand movements. The final results of this project should be a final report discussing the findings and a Matlab program for future use.

Methods like described above can be used to detect and recognize human hand movements, for example. An previous experimental system can be found in the video on the right where a Kinect is used to detect a human grasp. The model makes use of a trained low dimensional subspace of the human hand.

Contact for further information Kevin Luck


  • Programming skills in Matlab
  • Knowledge of Software Engineering
  • Some background of
    • matrix algebra
    • (continuous) probability distributions
  • Read and understand some specific scientific papers
  • Present a written draft of algorithms to implement
  • Design a small framework, such that the program is modularized and maintainable
  • Implement the algorithms
  • Implement a visualization
  • Extract the data from a dataset
  • Perform experiments given the dataset
  • Visualize and analyze the findings
  • Write a final report