Making hough transform more robust and precise

In this paper we present an extension of the Hough transform which enables us to reduce accumulator size without the loss of accuracy. The main idea is to use the output from the Hough transform as a set of possible hypotheses from which a selection procedure (based on the MDL principle) selects a set of candidates to describe the lines (curves). The selected lines are then subject to robust fitting procedure which improves the accuracy. The experiments on real data have shown that our algorithm is robust and produces highly accurate results. This approach can easily be extended to multiple curve type detection.