Extraction of tree trunk shape from a terrestrial laser scanning point cloud using the ransac algorithm
Kaja Kandare · Repozitorij Univerze v Ljubljani (Univerze v Lgubljani) · 2013
This BA thesis deals with a procedure for the extraction of a trunk shape and the calculation of the trunk parameters from a terrestrial laser point cloud of a tree. The tree trunk geometry has been simplified, so that it was modelled with a single and multiple cylinders. Two algorithms were employed for solving the problem: we used the RANSAC algorithm for the determination of the approximate values of the parameters and the extraction of the trunk shape, and the least squares method for the estimation of the parameters of the cylinder/cylinders with associated accuracies and for the calculation of the deviation between the model and the extracted point cloud. The process was carried out in different situations. The trunk was cut into partitions of different lengths, and for each partition, the parameters of the cylinder were calculated and the points which best fit the cylinder were extracted. Experimentally, we analysed one, seven, and fourteen slices. The results were different 3D models of the trunk, composed of a different number of smaller cylinders. The resulting 3D models of the trunk were visually and graphically compared with the corresponding extracted point cloud. We have proved that by increasing the number of slices, we improve the description of the trunk shape and determine the parameters of the trunk more accurately.