DetailseiteDetails

Indoor Location Recognition using Point Cloud Registration and Kinect Fusion

(3D Bild-Analyse und Synthese)
Betreuer: Anas Al-Nuaimi
Indoor location recognition remains to be a challenge. Established localization systems such as GPS do not work indoors. One approach to location recognition developed at our institute is based on matching query images visually to a database of geo-tagged images. This approach which is based on techniques from content-based image retrieval (CBIR) has shown very promising results. In this thesis, this approach is extended by replacing the database of images by a 3D point cloud and the query by a local scan of the geometry of the environment using the Xtion RGBD sensor. The location retrieval is achieved using shape matching algorithms. This is motivated by the increased proliferation of mobile 3D sensors that can capture appearance (RGB) and geometry (Depth: D) on the fly. Furthermore, state-of-the-art 3D reconstruction algorithms, such as KINFU, allow capturing an extensive query with fine shape details giving this approach an edge as compared to standard CBIR-based localization wherein the query is a view-dependent representation of the environment.
Keywords: Computer Vision, 3D Processing
Anforderungen: C++, matlab, interest in computer vision and 3D geometry