Demo from: visual-experiments.com

From Wikipedia: RANSAC is an abbreviation for "RANdom SAmple Consensus". It is an iterative method to estimate parameters of a mathematical model from a set of observed data which contains outliers.


In the case of a 2D line fitting, Ransac is very simple:

bestScore = Infinity
bestInliers = []
for a sufficient number of iterations :-)
	currentModel = estimate a line from 2 points randomly selected in dataset
	currentScore = 0
	currentInliers = []
	for all points in the dataset
		p = current point in the dataset
		currentError = distance(currentModel, p)
		if (currentError < threshold)
			currentScore += currentError
			currentScore += threshold
	if (currentScore < bestScore)
		bestScore   = currentScore
		bestModel   = currentModel
		bestInliers = currentInliers

return bestModel, bestInliers, bestScore

Given a set of 2d points you can use Ransac to estimate a fitting line. But the resulting line should only be estimated using inliers and not be contaminated by the outliers. Example of ransac iteration: bad model estimated | final ransac iteration if everything went well :-)

Live demo:

Source code: ransac.js (core Ransac + RobustLineFitting) available under MIT license.