Line-based initialization method for mobile augmented reality in aircraft assembly

Abstract

A fast line-based initialization method for mobile augmented reality in aircraft assembly is proposed in this paper. The challenge of recognition and pose estimation of aircraft structural parts in monocular images on mobile devices is mainly addressed. Rather than local feature points, straight lines are extracted and utilized here since they are more suitable for aircraft structural parts, which usually have a solid color and lack textures to generate invariant descriptors. The geometric relationship between 3D world lines and their projections on the camera image are built to estimate the relative 6-DOF pose including rotation and translation, respectively. The inertial sensors of mobile devices are used to provide partial rotation information directly, and the rest is obtained through a voting process. To shrink the search space and, therefore, improve the speed and robustness of initialization, a color-based ROI mask and a vertical lines extraction method are proposed. To determine the translation, the color mask is first used to roughly approximate the real pose and then a possible-correspondence map of lines is built by parallelism alignment combined with the rotation already obtained. This paper describes the details of our method and also shows its effectiveness with experiments. Moreover, the proposed method is especially suitable for but not limited to the aircraft assembly field.