Advanced Human Hand Model with Dynamic Constraints

In this thesis, we present the creation process of a virtual 3D hand model with sufficient accuracy and flexibility for the tracking of a human hand. The model consists of an underlying skeleton and a skin surface. The skeleton is implemented as a hierarchical joint structure with 17 joints and 26 degrees of freedom. We define an extra imaginary joint at the palm to model the movement of the CMC joint of the thumb. The skin is rendered from a point cloud joint by a triangular mesh. Our model has 46,140 points and 92,063 triangles. A set of constraints is also defined on motion of the hand at its joints. This helps in reducing the search space during tracking as poses disallowed by those constraints do not have to be taken into account. We used static and kinematic constraints on the hand and present a method for a future implementation of dynamic constraints. Detailed investigation of one kinematic constraint (θDIP = 2/3 θPIP ) defining a relationship between the PIP and the DIP joints of a finger was carried out. When used, this constraint reduces the number of degrees of freedom of the hand from 26 to 22. We confirm the validity of the constraint for natural hand motion and show that it may help tracking in difficult motions like a grip where different finger joints move at once. It also has potential to help track faster motion, possibly if used with other constraints or refined. We tested our model in the tracking framework developed by Mr. Desmond Chik, and compared its performance to a model from ETH Zurich which Desmond initially used to develop his algorithms. The tracking results show that using a hand model that is better fitted to the hand being tracked does not make significant improvement on the tracking although it could reduce the amount of work done by the tracker. This indicates that the tracking framework is robust enough to be used to track different hand shapes as long as the key dimensions of the hand to be tracked are fitted to the model. Good tracking results of the thumb movement validate the inclusion of the extra imaginary joint for the thumb CMC to help reproduce natural motion of the thumb. Swiss Federal Institute iv