Dimensionality Features Extraction Based-on Multi-scale Neighborhood of Multi-samples UAV Point Clouds

The growth in quantity and quality of collected point clouds has made inevitable to find methods of algorithmic interpretation of these data for any engineering analysis. To date, the most critical steps are still usually performed manually. The fully automated analysis of point clouds acquired from the most arising technologies, such as those generated by photogrammetry obtained from the increasing use of Unmanned Aerial Vehicles (UAVs), has therefore become a topic of great involvement. In order to describe the local geometry at a given point, the spatial distribution of the others 3D points within a local neighborhood was typically taken into account. From the analysis of the covariance matrix, it was thus possible to extract the geometric characteristics useful for identifying any linear, planar or scattering behaviors of the digital reconstruction. Hence, a suitable scalar dimensionality approach had to be planned to identify all the interpretable geometric behaviors of the points at various research scales. Subsequently, one can opt for a single representative scale or a multi-scale approach that serves the purpose of the work. The proposal of this work was to analyze the behavior of the scalar dimensional approach but varying the average Ground Sample Distance (GSD) values of UAV imagery of the same scenario. Encouraging results were obtained expressing how the scenario reproduces the same dimensional behavior irrespective of the acquisition conditions and an empirical finding between the scale parameter r and just the geometric resolution value.