Curvilinear Image Regions Detection: Applications to Mobile Robotics

Abstract

This paper proposes a novel approach for visual features detection, which is based on the presence of objects whose shape can be modelled using cylinders or generalized cylinders. These specific structures are commonly found on indoor and outdoor scenarios, and their image representations, the so-called curvilinear regions, automatically deform with changing viewpoint as to keep on covering identical physical parts of a scene. The method is based on Marr's visual theory that proposes that visual objects can be decomposed in generalized cylinders. Also, part of the method can be compared to the behavior of AOS neurons, placed in the caudal intraparietal sulcus, that respond when an elongated object is visualized. Our detector reliably finds the same curvilinear regions under different viewing conditions. Evaluation results are given to demonstrate the performance of the approach and its ability to be applied for visual features detection in a mobile robot navigation framework.

Publisher note

To access the full article, please see PDF.