Abstract

Atechnique capable of high performance, real-time detection of discrete straight line segments is described that allows major improvements to traditional alternative techniques such as the standard Hough transform. Such improved features result from the following three underlying principles: (a) the adoption of a precise mathematical-theoretical characterization of discrete lines and their mapping into parametric spaces; (b) the incorporation of adaptive parametric mapping that allows full congruency in the representation of discrete straight lines; and (c) the inclusion of a pre-processing stage that performs preliminary curve segmentation thus constraining the parameter space to be further investigated, which is done in a zoomed fashion. The high performance of the proposed technique has been corroborated, comparatively to three alternative techniques including the Hough transform, through a formal and quantitative performance assessment that is also discussed.