Initializing 3D model-based tracking of dice

The initialization process of a 3D model-based tracking method with an extremely fast checking time is presented. One initialized, the algorithm will execute a 3D model-based tracking with pose determination. The tracking method is a two-steps approach, a prediction-correction method. We are studying the possibility of finding the minimum group of interest points and a set of rules which form all the possible movements of a 3D tracked object. Thereby, the computation time-memory for the tracking task could be reduced. We consider rigid objects which can be approximately modeled by a convex polyhedral shape. Our approach works in monocular videos and where the position of the camera is fixed.