The thesis documents two approaches I have tried to implement computer vision based 3D hand gesture interaction. The more promising approach is selected to implement a first prototype for evaluation with users. Finally, the findings are summarized and future work is listed.
The introduction motivates spatial hand tracking by examining visionary interaction paradigms such as Minority Report:
To reconstruct 3D hand postures from stereo cameras, the first approach models epipolar geometry and triangulation rays:
The second approach evaluates machine learning classifiers and feature extraction pipelines for robust gesture recognition:
Finally, a unified object-oriented data model integrates raw sensor frames, tracked hand blobs, and recognized interaction events:
If you have feedback or questions please do not hestitate to contact me! Use the comment feature below to give other people the chance to engage into the discussion.