Efficient representation and generation of consistent 3D and 4D maps

点击翻译按钮获取中文摘要

The aim of P6 is the development of data structures and algorithms that can manage large amounts of 3D data. This should allow for both quick access to relevant parts of the data for the calculation of occupancy maps and the visualization and at the same time also support efficient insertion, deletion, and completion of the data. At the same time data recorded at different times should be identifiable persistently maintained in the system. The key to achieve this goal is on the one hand the application of out-of-core approaches and on the other hand the use of compression algorithms. Therefore, we will extend the developed compression methods in order to support time varying versions of the raw data by a persistent representation. The efficient upgradeability and the multi-resolution capability of the representation should be guaranteed.In addition methods for procedural description of surface structures of buildings will be investigated and exploited to improve the compression rates. As has already been demonstrated in experiments in phase I, current methods for geometry compression are not able to handle rough surfaces, such as plaster, clinker, or on finer scales also wallpapers, since these structures constitute a major part of the entropy of the data to be compressed. We aim for representing these structures by procedural noise models, rather than to compress them. Having found the relevant parameters of a suitable procedural noise model statistical analysis of the data allows for the determination of anomalies such as cracks or other irregularities in the masonry. Furthermore, a surface with the same statistical characteristic can be reproduced from the compressed smoothed underlying geometry and the procedural noise model. The third focus of the subproject will be on methods for robust and efficient symmetry detection on rough scales which serves several purposes. Firstly, it helps in the production of occupancy maps, which for obstacle avoidance (P2/P3) are required. On the other hand, it facilitates the action generation at the autonomous exploration (P8). In addition, they will also improve the geometry compression further, since self-similarity can be detected and used on coarse scales already. Finally, discovered self-similarities can also be used for meaningful supplement or completion of geometry data, which can, if necessary, avoid multiple overflights of an object. Typical examples are bays or towers that appear more than once in the overall building and usually have to be captured from several sides. To this end, the primitive-based reconstruction method for point clouds co-developed by one of the applicants will first be integrated into the reconstruction methods of P5 and second be extended to symmetric subparts that can be detected in the point cloud.