INFLUENCE OF VOXEL SIZE AND VOXEL CONNECTIVITY ON THE 3D MODELLING OF AUSTRALIAN HEATHLAND PARAMETERS

Abstract. Point clouds acquired through laser scanning techniques are applied in the three-dimensional modelling of vegetation. They provide the three-dimensional coordinates of geometric surfaces with attributes. However, raw point clouds are unstructured and do not provide semantic, geometric, or topological information about an object. Voxelisation is a method for structuring point clouds. It is a generalisation of point clouds and therefore the voxel size and the voxel neighbourhood play a critical role in the processing. This research explores the influence of voxelisation of point clouds acquired of heathland in Australia and how it influences the three-dimensional modelling and the representation of important heathland structure using different voxel sizes and voxel connectivities. Voxel sizes of 0.4 m, 0.6 m, 1.0 m, 1.2 m and 1.6 m with a voxel neighbourhood connectivity of 6, 18 and 26 are examined for three-dimensional modelling and segmentation of heathland vegetation in Australia. The results indicate that the choice of voxel size and the voxel connectivity influence the representation of important heathland parameters. A smaller voxel size of 0.4 m provides a detailed representation of mallee structure while the the processing time is longer compared to a larger voxel size. While a larger voxel size produces blobs while the processing speed is shorter. The results from the voxel neighbourhood connectivity represent a stronger voxel connectivity of 26-connected voxels suitable for heathland modelling rather than a 6-connected voxels.