This thesis addresses the problem of estimation of the camera poses with respect to a rigid object, which is equivalent to the problem of tridimensional registration of a moving rigid object before fixed cameras. Matching, tracking and 3D reconstruction of feature points by a stereoscopic vision setup allows the computation of the homogeneous transformation matrix linking two consecutive scene captures. Robustness to errors is provided by the scene rigidity constraint. Accumulation of error is compensated through loop detection in the calculated camera poses. Experimental results show the validity of the obtained camera poses.
展开▼