物体モデルからの生成画像と実画像の適合による二次元形状復元

抄録

This paper proposes a 2-D shape reconstruction method from multi-viewpoint images. It is based on a principle that the shape of an object model is changed so that the projections from the model match with the actual images. This image matching scheme realizes an object shape reconstruction with fusing two kinds of information, i.e., object ridge positions and surface orientations that reflect on edges and shading in images, respectively. Stereo, silhouette contours and shape from shading have the characteristics complement to each other, and the proposed method succeeds them in advantages such as robustness and accuracy. We consider a polygonal prism as a 2-D target object to simplify the problem. Polygon model are treated as a 2-D object model and deformed by a dynamic programming optimization under the condition that the light sources and the camera positions are known.