Outlier Detection for Factorization-Based Reconstruction from Perspective Images with Occlusions

B k contains filled B k (also R) constraint on R . intersection of many k filled R Outlier Detection Assumption: One dominant motion and random independent outliers Main Idea Minimal configurations of points in triples of images are su#cient to validate inliers reliably. =# simplified ransac: Not some large but some minimal support is su#cient. 1. Choose a triple of images and six points in them -# T 2. If consistent with enough correspondences Repeat 1 and 2 until there is enough inliers estimate reconstruction with occlusions [5] (outliers#occlusions) New Outlier Detection Algorithm 1. Initial set of inliers. Using s. 2. Reconstruction from inliers using [5]. 3. Check all data whether it is consistent with the reconstruction. In each column, p, of R: (a) Random triple of image points, P (b) If repr. error small Repeat (a) and (b) until the column is su#ciently sampled 4. Iteration of Steps 2 and 3 while any new inlier appeared "