Estimation of 3D real motion: a model based approach

A framework for recovering the pose and motion for articulated real objects is proposed. This method uses a generic 3D parametric model of the object and introduces a function of adequacy between images. This function is positive, it quantifies a concept of difference between two images. Minimizing the adequacy function allows to derive the solution. The minimization is achieved by using a generalized simplex-based algorithm which is easy to program, does not require derivatives and is robust to noise and occlusions. Experiments on synthetic and real video are presented in which the human hand is successfully tracked after recovering the initial configuration. Our method is both robust and generic. It could easily be applied to other shape and motion recovery problems.