Animation von Avataren bei physikalisch basierten Kleidungssimulationen

This Bachelor thesis deals with collision detection between clothing and moving avatars during a physically based simulation. A former thesis using distance fields to detect collisions is adjusted to work with dynamic situtations. A new file format to save animations has been specified. Generated animations may step into a loop to show longer motions without producing huge amounts of data. Collison detection is performed by interpolating the distance fields which hold the distance to the geometry of the avatar at a given time. These distances are precalculated for all keyframes of the animation. Elapsed simulation time and the frequency of keyframes is used to detect the current frames and their interpolation. In addition to the interpolation some special cases must be examined to avoid errors. So-called "belt"-particles are used to prevent the trousers from floating down through the influence of gravity. These particles are fixed relatively to the avatar using relative coordinates. "Belt"-particles are given an infinite mass to make them a fixed point to there neighbors. This also prevents extern forces from changing their velocity and consequently their position. So-called "critical" particles are also treated differently, because these particles might be pulled back into a state of collision by there neighbors after collision response. Internal spring forces between particles are adjusted so that only uncritical particles are moved after the collisionresponse. In addition all "critical" particles are moved along the direction of movement to reduce the number of collisions. This direction is determined by the difference of the jointpoints in the keyframes. To validate the above concept and test it's stabilty and performance different Java classes have been implemented and integrated into the existing "Virtual TryOn"-Prototyp