The clusterization tool detects objects using 2 parameters:
Separation distance: The minimum distance between point groupings. The higher this value is, the more probable it is that the software will merge nearby point groupings. A smaller separation value, however, may split an object into 2 or more clusters.
Minimum points: The minimum number of points for a grouping of points to be considered a cluster. The smaller this value is, the more probable it becomes that objects containing few points will be considered as clusters.
There is also an option for downsampling when detecting clusters. This option will speed up the detection process as fewer points will be analyzed during the detection. The downsample option keeps 1 point per cell and discards the rest of the points during the detection. The size of the cell is determined by the value inscribed for the “downsample size”. The higher this value is, the larger the cell size used; therefore, less points will be analyzed. A smaller downsample size will discard fewer points for the detection; therefore detecting clusters more accurately while taking more processing time.
The area to be analyzed must be considered before the detection takes place. The user can segment the point cloud using the fence tool or the clipping box; the clusterization tool will then only work on visible points. Also, classes can be shut down like the ground (which is highly recommended) so that points in theses classes will not be processed. The user may also create an extraction of an area within the point cloud. Detecting clusters within an extraction speeds up the detection process because of the reduced amount of points considered.
The "Small cluster to class" option allows to classify points that would appear in clusters with less points then the minimum required for them to be kept as such. It is usually done to classify small group of points as noise.