Scene understanding using omnidirectional images for indoor mobile robotic services

For a mobile robot equipped with a camera, the vanishing points detection is an important machine vision task necessary for various applications, such as mobile robot navigation and camera calibration. In this paper, a novel algorithm to detect the vanishing point in various indoor environments using an omnidirectional camera is proposed. The proposed algorithm contains three steps. In the first step, the Local Dominant Orientation Signature (LDOS) descriptor is applied to extract structural features in an image. In the second step, many unreliable features are eliminated. Thirdly, the Radial Basis Function (RBF) is used to generate an orientation field of the whole image. Finally, the location of the vanishing point is estimated. Experimental results show that the proposed technique is able to robustly and accurately estimate the vanishing point for a variety of orientations of the camera mounting. Experimental results in diverse environments have demonstrated that the proposed technique is able to accurately and robustly detect the vanishing point.