A Sparse 3D Model Reconstruction Algorithm Integrating Improved DAISY and MLESAC-A

Sparse 3D model reconstruction is very important in many fields such as robot autonomous mobile, unmanned driving, AV/VR and so on. The feature point matching and camera pose estimation are the key steps, which affect the reconstruction accuracy of sparse 3D model. In this paper, an improved feature descriptor, which mixes local and global features in DAISY, is used for feature matching. And the MLESAC-A, which involves pre-inspection step, is used to estimate the initial camera pose. The final camera pose and the 3D model are optimized by minimizing the total re-projection error. The experimental results show that the reconstruction accuracy of 3D sparse model by the proposed algorithm is higher than that of traditional structure form motion algorithm based on SIFT.