MLESAC: A New Robust Estimator with Application to Estimating Image Geometry

By Philip H. S. Torr, Andrew Zisserman. Computer Vision and Image Understanding, 2000

2,679 citations recorded by the source corpus.

Abstract A new method is presented for robustly estimating multiple view relations from point correspondences. The method comprises two parts. The first is a new robust estimator MLESAC which is a generalization of the RANSAC estimator. It adopts the same sampling strategy as RANSAC to generate putative solutions, but chooses the solution that maximizes the likelihood rather than just the number of inliers. The second part of the algorithm is a general purpose method for automatically parameterizing these relations, using the output of MLESAC. A difficulty with multiview image relations is that there are often nonlinear constraints between the parameters, making optimization a difficult task. The parameterization method overcomes the difficulty of nonlinear constraints and conducts a constrained optimization. The method is general and its use is illustrated for the estimation of fundamental matrices, image–image homographies, and quadratic transformations. Results are given for both synthetic and real images. It is demonstrated that the method gives results equal or superior to those of previous approaches.

Subfields: Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/40d30ade023d671c0f4e41e7045f3e59db44edcc

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

Citations in the corpus

MAGSAC++, a Fast, Reliable and Accurate Robust Estimator (2019, methodology): “The compared methods are RANSAC [5], LMedS [22], MSAC [28], GC-RANSAC [1], MAGSAC [2], and MAGSAC++.”
A review of multimodal image matching: Methods and applications (2021, methodology): “Many improvements based on RANSAC mainly focus on verifying the model quality, such as using a maximum likelihood procedure in MLESAC [307,308] or modifying the sampling strategy as inspired from the specific properties of inliers, such as spatially consistency-based [312], by using a grouping strat”
Structure from motion photogrammetry in physical geography (2016, methodology): “Other similar ‘hypothesise-and-test’ frameworks may be implemented at this stage, such as the Maximum Likelihood Estimation Sample Consensus (Torr and Zisserman, 2000) or the Hough transform (Ballard and Brown, 1982).”
Robust monocular SLAM in dynamic environments (2013, background): “There are already several robust estimation algorithms [13, 38, 7, 6, 23, 5] have been proposed.”
The Visual Object Tracking VOT2014 Challenge Results (2014, positive): “The FoF tracker computes a sparse Lucas Kanade flow [41] and uses MLESAC [55] with similarity transform to predict the target position.”
Image Mosaicing and Super-resolution (2004, methodology): “The algorithm used for this purpose is Torr and Zisserman’s MSAC (M-estimator SAmple Consensus) algorithm [160], a variation of the previously described RANSAC algorithm.”
MUlti-Store Tracker (MUSTer): A cognitive psychology inspired approach to object tracking (2015, methodology): “In our implementation, we use the robust MLESAC estimator [15, 45], which works by maximizing the likelihood rather than just the number of inliers.”
Bayesian Model Estimation and Selection for Epipolar Geometry and Generic Manifold Fitting (2002, methodology): “The generic method of using random samples to maximize the posterior (or an approximation of it) followed by gradient descent is dubbed MAPSAC maximum a posteriori sample consensus to distinguish it from the author’s previous work MLESAC which was purely a maximum likelihood formulation.”
Robust Point Matching via Vector Field Consensus (2014, background): “Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org.”
The structure-from-motion reconstruction pipeline - a survey with focus on short image sequences (2010, neutral): “The random sample consensus [13] has received many improvements such as MSAC and MLESAC [43], Locally Optimized RANSAC (LO-RANSAC) [8, 9], PROgressive ranSAC (PROSAC) [7], and more recently QDEGSAC [16] to detect quasi degenerate cases.”
Performance Evaluation of RANSAC Family (2009, methodology): “MSAC (M-estimator SAC) [30] adopts bounded loss of RANSAC as follows: Loss(e) = { e2 | e |< c c2 otherwise .”
Close-range scene segmentation and reconstruction of 3D point cloud maps for mobile manipulation in domestic environments (2009, background): “The right part of Figure 5 presents the remaining point clusters Ci after the previously mentioned segmentation steps for a given table from the kitchen dataset shown in the left part of the same figure.”
Matching with PROSAC - progressive sample consensus (2005, methodology): “Tordoff and Murray [14] combine the MLESAC [15] algorithm with non-uniform (guided) sampling of correspondences.”