Description
Acquisition and processing of geometric data from the Earth surface are complex processes that used to be considered difficult, slow and expensive. However, modern technologies enabled development of devices capable of fast and data capturing. The focus has therefore been moved to data storage and processing. LiDAR (Light Detection and Ranging) scanners can capture up to 24 points in a square metre, with the acquisition speed as high as 200.000 points per second. The results of this are huge amounts of geometric data – 3D points with attached specific data that exceed the storage capacities of a computer system. Their storage and processing require special approaches. The development of software for huge datasets processing hardly follows capabilities of data capture devices. This implies many challenges and problems in the field of geometric data processing. In the proposed project, we deal with some of them in close cooperation with final users. These problems are represented in greater detail in continuation. The companies Igea d.o.o., X-Lab d.o.o., Dat-Con d.o.o. and Geoin d.o.o., which partially fund the proposed project, are either data providers or distributors. The expected solutions will directly support the quality of their services. To achieve the set goals, thorough theoretical knowledge of the field, new theoretical solutions and practical experiences are necessary. Proven excellence of our previous research results and successful transfers to practice provide a good starting point.
The project activities will run in three main directions, where the particular solutions will supplement each other. Visualization of geometric data represents the most natural way for results adequacy evaluation. Because of huge amounts of data, a LOD-based (level of details) approach will be used. Since we deal with unstructured data, we will focus on point-based rendering (PBR) method and graphic processors utilization. Existing PBR visualization of LiDAR data still does not reach the satisfactory quality and, therefore, improvements are extremely important from theoretical and also practical point of view.
Data acquired from capture devices are stored in files. Owing to their size, archiving is exposed to enormous problems and the data transfer through internet is practically impossible. General-purpose data compression algorithms only partially facilitate this problem. The solution lies in design of domain-specific algorithms for compression of geometric data. We intend to implement data compression and decompression algorithms as streaming algorithms with prediction, also suitable for hardware implementations. Here we will rely upon the modern concepts of high-level synthesis and corresponding platforms from the field of highly efficient reconfigurable computer systems.
Feature detection (segmentation) in unstructured geometric data is vital for numerous applications. Known approaches often prove inefficient and numerically intensive. In the proposed project, we intend to develop algorithms which utilize fast geometric techniques to early eliminate those data that certainly do not represent a desired feature. In this way, we will significantly reduce data amounts and enable use of known approaches on the remaining data. This will considerably improve the method efficiency.
The improved visualization, domain-specific realization of data compression algorithms, their hardware implementation and innovative approaches to data segmentation algorithms design will be incorporated into a framework of user tools intended for direct practical use. The achieved theoretical results will be published in scientific publications. An important project goal is also the LiDAR data incorporation into the existing distributed geographic information systems (GIS). The accuracy and actuality of visualized data will be considerably improved and, consequently, the usability of such GIS-based applications in everyday practice will be intensified.