3D Model Reconstruction by Multiple-Views Feature Based Matching and Silhouette Information during Time

This work attempts to find a way out to existing problems of three-dimensional model reconstruction like uncertain stereo matching in low contrast regions; the poor color adjustment between cameras; occlusion due to using the limited number of cameras; camera visibility test; poor color consistency of object due to shading. A new approach is proposed to extract the visual hull of rigid object using its silhouettes captured by multi cameras during time. A robust feature based multiple views matching is used to find the precise position of some sparse feature points belonging the surface of object. These feature points are then used to estimate motion parameters in the next sequences. Combining the calibration matrix of each camera and the motion matrix for any sequence makes the new virtual cameras. These constructed cameras increase the number of silhouettes. Cone intersection method is employed to extract the bounding edges visual hull from the silhouettes. Finally, after reconstruction of the visual hull geometry, the color mapping of the 3d points is completed by projecting them to the visible cameras.

Keywords:

3D model reconstruction , Visual Hull , Silhouette , Feature based stereo matching , Feature based motion estimation