Combining the feature based stereo matching, motion and silhouette to reconstruct visual hull

This paper proposes a new approach to reconstruct the visual hull of rigid object using its silhouettes captured by virtual cameras during time. The main idea of this work is to improve the quality of reconstruction by combing the advantages of different methods. Structure from silhouette, structure from motion and depth from stereo are popular methods in 3D reconstruction. But all of these methods suffer from some drawbacks. For example stereo matching fails to extract depth in low contrast regions but it works well in edge regions. Structure from silhouette can be employed to extract 3D shape of texture less object but it needs a lot of cameras to do fine. In this paper a robust feature based stereo matching of multi camera images is used to find the exact place of some sparse feature points on the surface of object. These 3D feature points are then employed to estimate six motion parameters in the next sequence. New virtual cameras are constructed by multiplying the calibration matrix to motion matrix in each frame. So, a lot of cameras can be constructed for the moving object during time. After all, the cone intersection method is used to extract the bounding edges visual hull from all the silhouettes captured by virtual cameras.