3D Point Cloud Optimization Based on a Random Walk Model Using SBA

In this paper, we propose a new algorithm for optimizing coarse 3D point cloud obtained from multiview images. We start firstly with the 3D geometry acquired from images taken from multiviews by structure from motion (SFM) approach. Then, the 3D point data is regarded as the input of the sparse bundle adjustment (SBA) technique in a RANSAC framework. A random walk model is proposed to relocate the back projections of 3D points which are also one of the inputs of the optimization algorithm, so as to improve the optimization results further. Experimental results in the end of the paper prove the effectiveness of the proposed approach.