Model Driven 3-D Shape Reconstruction by Matching Synthesized Images with Actual Images

In this paper we propose a 3-D shape reconstruction method from multi-viewpoint images under the condition that the parameters such as poses of light sources and a moving camera are known. It is based on a principle of a model-synthesized image matching in which the shape of an object model is changed so that the model-synthesized images are matched with the multi-viewpoint input images. A 3-D target object is modeled as a triangular polygon. For matching, it is deformed by moving the vertices with a non-linear least squares method, and, if necessary, it is changed by dividing a triangle into some triangles and by merging some triangles to a triangle.