LIDAR (Laser Imaging Detection and Ranging) detects a rock mass surface as a point cloud, and threedimensional configurations of the rock mass can be obtained from the point cloud. In previous studies, algorithms to estimate discontinuities from a point cloud have been developed. In those algorithms, it is necessary to determine geological parameters in advance. DiAna(Discontinuity Analysis) is a Matlab tool which was developed for geostructural analysis of rock mass discontinuities. It is a semi-automatic tool. DiAna segments a point cloud into bounding boxes to estimate the surface of a rock mass. However, an expert's skills necessary to determine the appropriate size of the bounding boxes for DiAna. We developed the VBS (Variable-Box Segmentation) algorithm to determine the appropriate box size depending on the location of the point cloud and to estimate the surface of a rock mass. The VBS algorithm consists principally of three processes: large box segmentation, small box segmentation, and merging. The small boxes are merged to obtain an appropriate box size. The surface of the rock mass is estimated using the point cloud in the box. The performance of the VBS algorithms was evaluated using point clouds obtained by a geological survey. For evaluation, we estimated reference rock mass surfaces manually using the point cloud and geological sketches by geologists. Similarities among the respective reference surfaces and the surfaces estimated using the VBS algorithm were measured. Similarities among the respective reference surfaces and the surfaces estimated using the DiAna algorithm were also measured. The similarities among them were compared using standard competition ranking. The results of comparison showed that the VBS algorithm estimated planes more accurately for the reference planes than the DiAna algorithm. Therefore, the VBS algorithm determines appropriate box sizes automatically depending on the location of the point cloud and estimates the surface appropriately.
