Fig. 1.

A possible RANSAC output where the desired model is of the first power (i.e., straight line). The algorithm assumes that due to internal noise, samples will not be exactly on the model but within a normal distribution around it. Conceptually, this variance forms a “strip” where all samples that lay within its boundaries are influenced by internal noise only. Samples within a “strip” are defined as model-compatible. Samples outside the “strip” are defined as model-incompatible