A probabilistic framework for the hough transform and least squares pose estimation

The Hough transform and least squares pose estimation are usually considered as unrelated methods based on different assumptions. This paper presents a unified perspective of both approaches, in a probabilistic framework. It is shown that both methods compute maximum likelihood estimates of the object pose, based on different probability distributions. The main properties of the algorithms are discussed and their performance is characterized by Monte Carlo tests. This methodology can be extended to other shape representation algorithms.