Important approach to vanishing points detection based on simple image geometry and new accumulation space

Abstract

Vanishing points detection is a basic problem in computer vision. It is widely studied in various areas of research. Self-calibration, 3D reconstruction and robotics are among the important which use vanishing points. Several methods exist already; they are complex because of difficult mathematical steps. As a result, they are expensive in terms of time and memory resources. In this paper, we present a new method of vanishing points detection. We evaluate reliability and accuracy of obtained results using a new accumulation space based on Thales theorem. The developed algorithm in this article brings us a points cloud from image using simple 2D geometry. And then, we handle it in order to detect vanishing points. Obtained results from simple and complex images shows clearly that our proposed method is very easy to be developed and implemented thanks to a reduced number of simplified method instructions. Unlike the majority of existing methods which use classical space of accumulation based on lines intersections, our proposed method brings more accurate results, reduced accumulator size and efficient accumulation step. Such method can be used in various other processes of computer vision if an easy fast way thanks to its simplicity.