A registration method to create a complete model of a 3-D object is presented in this thesis. If one needs a complete model of an object, the following steps are necessary :
1. data acquisition,
2. registration between views,
3. integration of views.
The first problem that we need to solve is to find a transformation between two different views. Previous approaches have either assumed this transformation to be known (which is extremely difficult for a complete model), or compute it with a feature matching (which is not accurate enough for integration). In this thesis, we propose an approach which works on range data directly, and registers successive views. The method consists of two stages : an initial estimation and a final refinement of the transformation. Without an exact feature matching, an initial estimation of the transformation is obtained by minimizing a distance between a point in one view and a tangent plane in another view. For a final refinement of the transformation, we propose an iterative method which is composed of predicting corresponding points and updating the transformation matrix iteratively. For creating a model of a 3-D object, we integrate these multiple registered range views into a composite 3-D model. Such models can then be used in object recognition, pose estimation or inspection tasks. If the object of interests has been precisely designed, then such model exists in the form of CAD model. In many applications, however, it is not possible or not practical to have access to such CAD models, and we need to built models from the physical object. We give the details of the registration method and modeling procedure, and illustrate them on real range images of complex objects.
