This web site is no longer maintained and the content may be outdated.
Please visit for up-to-date information.
No upcoming events...

Home / Graduate / M.S. Theses Completed
  Ayşe Naz Erkan, 2004    

Thesis Title

Model based three dimensional hand posture recognition for hand tracking


This study focuses on model based hand posture recognition, which is the acquisition of the static hand pose information. Here, we have applied geometric modeling with a simple 3D hand model constructed with basic geometric shapes. Our similarity measure is the non-overlapping area of silhouette of the model and the images acquired from the camera. This measure is optimized so as to estimate the best matches using two search methods, the Genetic Algorithm and the Downhill Simplex Method. In order to resolve the occlusion problem, the biomechanical constraints of finger motion are integrated to the search algoritm. This helps reduce the search space and eliminate convergence to local minima to some extent. A more significant remedy to attack ambiguities is the use of multiple camera systems. The proposed system uses two cameras and is scalable to any desired number of viewpoints. Our results reveal that even with a very simple model, 3D hand pose reconstruction can be achieved in occluded poses. The proposed system promises better results with more detailed and accurate hand models. Our experiments also demostrate that GA-DS hybrid algorithm outperforms GA in the constrained multidimensional global optimization of hand model parameters.
Boğaziçi University Department of Computer Engineering
Address: 34342 Bebek, Istanbul, TURKEY
Phone: +90 212 359 4523-24 Fax: +90 212 287 2461
general information:   webmaster: