This paper describes complex networks in regards to their fractal characteristic. Firstly, a short background concerning fractal dimensionality based on box-counting is considered, while in the continuation its application on spatially embedded networks is proposed using rasterization at multiple scales. Finally, the obtained results are presented.

B.03 Paper at an international scientific conference

COBISS.SI-ID: 18157078In this Doctoral dissertation, we introduce a new algorithm for the classification of vegetation points within LiDAR data. The classification procedure can be outlined with two steps: An analysis of the distribution of points and an analysis of the context in which they are located. Vegetation points are, namely, characterised by their non-linear distributions and they can, therefore, be recognised effciently in relation to the large plane fitting errors. The classification is improved further by introducing three contextual filters, which deal with attached objects (e.g. walls, chimneys, balconies), overgrown vegetation and small objects (e.g. vehicles, fences, statues). We have shown that the proposed algorithm is able to classify vegetation despite of the different vegetation types (deciduous and coniferous), enviroments (mountain, forested,urban), and different levels of leaf conditions. An average F1 score of 97.9% was achieved for non-urban areas and 91% for urban areas, which are, in terms of classification difficulty, considered to be more difficult. The algorithm uses only the geometrical information of points, which presents an advantage when compared with the methods which rely on point clouds with high point density and reliability (or the presence) of other information. The parameter sensitivity analysis has also shown that the three used input parameters are stable and robust. They, therefore, provide an efficient way to regulate the ratio between completeness and correctness.

D.09 Tutoring for postgraduate students

COBISS.SI-ID: 20461590This invention relates to the compression of 3D geometric meshes and point cloud data. The method’s embodiment consists of binary voxelizator, slice decompositor, chain code encoder, and entropy encoder. The binary voxelizator is responsible for voxelization of 3D points from 3D mesh vertices or point cloud data, where the resulting voxels within 3D grid contain a binary scalar value (i.e. empty or occupied). The compression is near-lossless, because of the quantization induced by the voxelization. Furthermore, only the compression of geometry is considered. The voxel 3D grid is then divided into 2D slices having a thickness of one voxel. The slices represent 2D raster grids, and are perpendicular to the coordinate axis of the shortest bounding box side of the considered 3D mesh or point cloud. Each slice contains one or more 2D segments consisting of occupied pixels. The segments are preprocessed before they are encoded by the chain codes. Optionally, thinning is applied on each segment in order to generate contours with the width of one pixel. The contours are then encoded by a chain code algorithm. The starting positions (x and y coordinates) of each contour are coded as an nearest offset to either the grid’s center of the 2D slice, or one of the previously coded segments, or from one of the grid’s four corners. The starting position is then codded with the variable length coding (VLC) scheme. The resulting chain codes from different segments are then further compressed by using entropy coding. The embodiment of the decompression is done by entropy-based decoding, chain code decoding into 2D segments, constructing a 3D grid out of decoded slices, and extraction of points from voxels. 3D points denoting extracted voxels’ centers positions are smoothed by average smoothing over each point’s local neighborhood. Optionally a reconstruction algorithm is applied in order to obtain an approximation of the initial 3D mesh.

F.32 International patent

COBISS.SI-ID: 20741398This invention considers a methods and apparatus for the lossless progressive compression of LAS files, obtained by the airborne Light detection and ranging (LiDAR) scanning. The procedure presented by this invention consists of the several steps: the determination of bounding box, dividing the bounding box into the 2D grid, defining the levels of grid lines, defining the level of point, distributing the point into the corresponding array, and storing the arrays by using compression method. The method performs the arrangements of the LiDAR data in the hierarchical levels without preprocessing in one pass due to the expected large sizes of LAS files.

F.33 Slovenian patent

COBISS.SI-ID: 19144214