論文

Simultaneous localization and mapping (SLAM) is a technique to simultaneously perform mapping of environments and localization of a camera in real-time. Most existing monocular vision based SLAM techniques use point features as landmarks. However, images of artificial environments with little texture often contain many line segments, whereas few point features can be localized in such a scene. We propose here a realtime line-based SLAM system, and a novel method for describing the features of line segments (LEHF:Line-based Eight-directional Histogram Feature) in order to establish correct 2D and 3D line correspondences (2D-3D correspondences). LEHF is a fast and efficient way of describing features of line segments, which are detected by the line segment detector (LSD) method. The line-based orthogonal iteration (LBOI) method and the RANSAC algorithm are applied for the camera pose estimation. We conducted an experiment in order to test our SLAM system in a desktop environment and to perform augmented reality (AR). Moreover our SLAM system was evaluated by synthetic data.

We present a method for overlaying a texture onto a non-rigid surface using a commodity depth camera. The depth cameras are able to capture 3-D data of a surface in real-time, and have several advantages compared with methods using only standard color cameras. However, it is not easy to register a 3-D deformable mesh to a point cloud of the non-rigid surface while keeping its geometrical topology. In order to solve this problem, our method starts by learning many representative meshes to generate surface deformation models. Then, while capturing 3-D data, we register a feasible 3-D mesh to the target surface and overlay a template texture onto the registered mesh. Even if the depth data are noisy or sparse, the learning-based method provides us with a smooth surface mesh. In addition, our method can be applied to real-time applications. In our experiments, we show some augmented reality results of texture overlay onto a non-textured T-shirt.

In this paper, we present a new online video-based rendering (VBR) method that creates new views of a scene from uncalibrated cameras. Our method does not require information about the cameras intrinsic parameters. For obtaining a geometrical relation among the cameras, we use projective grid space (PGS) which is 3D space defined by epipolar geometry between two basis cameras. The other cameras are registered to the same 3D space by trifocal tensors between these basis cameras. We simultaneously reconstruct and render novel view using our proposed plane-sweep algorithm in PGS. To achieve real-time performance, we implemented the proposed algorithm in graphics processing unit (GPU). We succeed to create novel view images in real-time from uncalibrated cameras and the results show the efficiency of our proposed method.