Detecting 3-D Parallel Lines for Perceptual Organization

Abstract

This paper describes a new algorithm to simultaneously detect and classify straight lines according to their orientation in 3-D. The fundamental assumption is that the most “interesting” lines in a 3-D scene have orientations which fall into a few precisely defined categories. The algorithm we propose uses this assumption to extract the projection of straight edges from the image and to determine the most likely corresponding orientation in the 3-D scene. The extracted 2-D line segments are therefore “perceptually” grouped according to their orientation in 3-D. Instead of extracting all the line segments from the image before grouping them by orientation, we use the orientation data at the lowest image processing level, and detect segments separately for each predefined 3-D orientation. A strong emphasis is placed on real-world applications and very fast processing with conventional hardware.