NAPSAC: high noise, high dimensional model parameterisation - it's in the bag
Darren R. Myatt, Philip H. S. Torr, Slawomir J. Nasuto, J. M. Bishop, R. Craddock · 2002
Abstract. The most effective algorithms for model parameterisation in the presence of high noise, such as RANSAC, MINPRAN and Least Median Squares, use random sampling of data points to instantiate model hypotheses. However, their performance degrades in higher dimensionality due to the exponentially decreasing probability of sampling a set of inliers. It is suggested that biasing this random selection towards clusters in multi-dimensional space may be preferable. Based on this premise, the NAPSAC (N Adjacent Points Sample Consensus) algorithm is derived and its performance in high dimensionality is shown to be superior to RANSAC on a hyperplane fitting problem. 1.