Parallel line-based structure from motion by using omnidirectional camera in textureless scene

Abstract

In this paper, we propose a reconstruction method for a 3D structure using sequential omnidirectional images in an artificial environment. The proposed method is fundamentally categorized into the Structure from Motion (SfM) technique. The conventional point-based SfM using a standard camera is, however, likely to fail to recover a 3D structure in an artificial and textureless environment such as a corridor. To tackle this problem, the proposed technique uses an omnidirectional camera and line-based SfM. Line features, such as a borderline of a wall and a floor or a window frame, are easy to discern in an artificial environment comparing point features, even in a textureless scene. In addition, an omnidirectional camera can track features for a long period because of its wide field-of-view. Extracted line features in an artificial environment are often mutually parallel. Parallel lines provide valuable constraints for camera movement estimation. Directions and locations of lines are estimated simultaneously with 3D camera movements. A 3D model of the environment is constructed from measurement results of lines and edge points. Experimental results show the effectiveness of our proposed method.