A comparison of terrain classification using local feature measurements of 3-dimensional colour point-cloud data

This paper describes an examination of local feature measures for the classification of 3D range data collected using low cost sensors mounted on an unmanned air vehicle. Specifically this paper proposes and assesses the use of local metrics such as curvature, Zernike moments, colour and SPIN representations used in 3D object recognition for the labelling of terrain structures. The results of applying these features to sample data are presented. The classification scores achieved demonstrate that structures including vegetation, ditches and similar earthworks can be classified using such local feature estimates. © 2009 IEEE.Danny Gibbins and Leszek Swierkowsk