Optimizing Computational Performance for Real-time Mapping with Airborne Laser Scanning

This paper focuses on evaluating and optimizing the computation procedures for achieving real-time mapping with modern Airborne Laser Scanning (ALS) instrumentation. The computation steps under consideration include real-time point cloud generation at full resolution, strip-wise analyses of scanning density and coverage, error propagation, ground classification and digital terrain and surface model production and DTM quality map. Different approaches for each mentioned step are implemented within a custom system and evaluated under real mapping scenarios for speed and correctness. The investigation reveals that real-time generation of laser point cloud (georeferencing) is feasible with conventional laptop for scanning rates up to 200 kHz. Further it is shown that the analyses of point cloud density and coverage are less computationally demanding, so is the calculation of digital surface model (DSM) and generation of corresponding hillshade image. Therefore, these tasks can be completed shortly after each flight line. On the other hand, the most computationally demanding tasks are identified as point-cloud classification required for DTM production as well as the error propagation. The latter tasks are therefore examined for optimizing their execution speed while maintaining correctness. The retained solutions are presented in the paper together with recommendation for further development.