A Research On Image-Based Virtual Environment Navigation Technology

Image-based rendering (IBR) has attracted a lot of research interest recently. In contrast with the conventional geometry-based approaches of using 3-D models for creating new virtual views, IBR techniques use photo-realistic and render directly from these original images based on pixel interpolation or reprojection. These new techniques have many advantages over traditional geometry-based rendering, such as high quality rendering, speed independent of scene complexity and only relies on processor speed. This thesis is focused on some algorithms such as corner detection, cylindrical panoramic mosaic and image matching algorithm.Corner detection has been shown to be extremely useful in many computer vision applications. This thesis describes some robust methods for image corner detection. We propose two simple and fast algorithms to detect comers. The first method based on the geometric character of the corners. The corner points of the image are defined as the junction point of two or more straight line edges. The second method is a two-stage detection method, coarse candidates of corners are extracted using Harris detector, and then SUSAN principle is used to remove false corners. Experimental results demonstrate that those algorithms proposed in this thesis is effective and valuable.Panoramic image has important implications in robotics, computer vision, surveillance and virtual reality. We propose a new method to mosaic images. The corners are matched by correlation-based method after they are detected from images. Then a circulation process is used to remove the mismatched points and a credible correspondent points volume are found out. Without any manual selection of features, it can mosaic panoramic image automatically and quickly.Furthermore, we construct a virtual environment navigation system, which employs panoramic image and spidery mesh model of tour into the picture. With the path of the viewpoint is known, we realized two touring behaviors, namely, looking around into the panoramic image and moving in one direction into image sequence. Experimental results demonstrate that the navigation system is effective and applicable.