Robust Multi-View 3D Reconstruction Of Human Body

3D reconstruction of the human body is a very important research topic in computer graphics, especially in 3D vision field, meanwhile it is a challenging engineering research area. This paper proposes a robust multi-view 3D reconstruction method for human body. We develop multi-sequence alignment system and depth optimizer, so that we could align multiple sequences and optimize the depth map. The refine model is reconstructed after the following process such as camera tracking, depth recovery, sequence alignment and depth optimization template-based. The proposed method can eliminate the depth error and exclude background information. By combining the effectiveness of depth information, our tiered adaptive RANSAC-based iterative strategy could solve similarity transformation between sequences efficiently. In order to settle the problem of poor depth quality for texture-less region, the depth map is limited to the vicinity of the template. In view of this, we propose a new alignment method with which the model is aligned by main direction. The main direction is computed by decomposing human body model. The model to be aligned is deformed to better fit the target also. The new alignment method is proved to be effective in variety of human body models. The experience shows that the proposed method could realize smooth, ergonomic human body reconstruction.