Detección Eficiente de Elipses en Imágenes. Aplicación al Posicionamiento 3D de un Robot Industrial

Abstract

In this paper, we present a ellipse detection algorithm developed to measure the 3-D position of a circular feature in a robotic application. The algorithm uses a RANSAC stochastic procedure whose efficiency has been significantly improved, substituting the random sampling with a guided sampling on the curve segments in the image. The contours of the image are first split analyzing their curvature. Then the curve segments are sorted according to their likelihood to be part of the ellipse. We have used the length as a prior indicator of this likelihood. The RANSAC algorithm starts considering only the longer curve segments whilst shorter curve segments are progressively incorporated. This strategy notably diminishes the amount of samples needed and makes the algorithm suitable for real time.