Print Email Facebook Twitter Point Clouds in a Database: Data Management within an Engineering Company Title Point Clouds in a Database: Data Management within an Engineering Company Author Wijga-Hoefsloot, M.E. Contributor Zlatanova, S. (mentor) Faculty OTB Research Institute for the Built Environment Department GIS-Technology Programme Geomatics Date 2012-09-27 Abstract With laser scanning (including laser altimetry and multi-beam echo sounding), many data points, called point clouds, are measured. The interest in point clouds is increasing. Depending on the laser scanner, the measurement set-up and the technique, the number of points in one dataset can vary from a few hundred points to over a billion points. The data volume of a laser survey mission can easily reach Gigabyte or even Terabyte level. It is a problem to handle these voluminous datasets efficiently. One way to manage the data is to partition the data with tiles or grids and then store the data in each tile or grid as a single file in text or binary format; the large volume of data is divided into separate files of a reasonable size. As an alternative, the data can be stored in a database management system (DBMS); each point in a single record based on standard data types. Since the data stored can be accessed in single point level, it is easy to perform queries and analysis on the data server, and only the qualified points are returned to the user. Since the data is spatial data, it has a location; data can be stored based on a spatial data type in a spatial DBMS, spatial objects can be clustered and spatially indexed to limit searches, with the result that spatial queries can be performed. Storing and retrieving massive datasets as single point records (one point per record) results in storing and retrieving lots of records, but fast retrieval become questionable, as there is overhead per record and because of the number of records to be retrieved. As Oracle is the market leader in the development of DBMS, this research is limited primarily to Oracle. This leads directly to the research question of this thesis: What is the best design for a data model to store large point clouds in an Oracle DBMS, such that it is generally accessible by spatial applications, that all attributes are preserved and that performance is optimised? Analysis A solution has been found by reducing the number of records by clustering nearby points without loss of information. If points are clustered, associated attributes have to be clustered as well. It has been decided how to record these attributes with the points. The Region Quadtree is implemented to grid the data in logical parts. The Morton space-filling curve is implemented to cluster the data. Hilbert space-filling curve would even be better. The points have been stored using a well-known data type called SDO_GEOMETRY (GTYPE 3005, POINTCLUSTER), SDO_GEOMETRY is a standard spatial data type implemented in Oracle. Attributes have been stored in one or more arrays (VARRAY), because ordinates are also stored in a VARRAY. This method is been compared to the newer SDO_PC data type. Conclusion The presented method using SDO_GEOMETRY (GTYPE 3005, POINTCLUSTER) performs optimal for large point cloud datasets in terms of performance and the points can generally be accessed by spatial applications. The alternative method using SDO_PC is not suitable for the conditions as stipulated in the research question. Because the SDO_PC data type is newly implemented in Oracle, not generally readable by spatial applications, not (yet) fully supported by all vendors and required storage space is double the amount. Subject OraclePoint CloudSDO_GEOMETRYmulti pointspace-filling curve To reference this document use: http://resolver.tudelft.nl/uuid:58266d2c-330c-4493-a03a-d8b180eb810d Part of collection Student theses Document type master thesis Rights (c) 2012 Wijga-Hoefsloot, M.E. Files PDF Master_thesis_-_report.pdf 2.1 MB Close viewer /islandora/object/uuid:58266d2c-330c-4493-a03a-d8b180eb810d/datastream/OBJ/view