3D points obtained at a given frame need to be registrated to the current model before it is updated. A RANSAC approach is to be used to estimate the transformation. Reconstructed 3D points from the current model and those in the current frame are matched again using the local appearances. Then three points are randomly selected from the list of best matches and a global transformation is computed from their coordinates in the current frame and the previous one. After that, the system verifies if the transformation fits other good matches. If not, the system takes three other randomly selected points and repeats the process until a good result is found. If the approach does not converge to a good result after a number of trials, the best result is kept. Because an animal is a deformable object, the system needs to detect if some parts of the animal have moved differently. The approach taken is based on the skeletal signature of the 3D object. If the system has at least 3 points from a skeleton branch in the list of good matches, it can compute the specific motion of that branch. At the end of this step, the new 3D points (those that are not in the list of good matches) are added to the 3D model of the object and the points from the deformed parts are updated.