Beginning in 2014, some grouping tools were introduced to TerraScan. Those tools applied to mobile LIDAR applications as they were aimed at classifying moving cars and trees. In late 2016, additional grouping tools were introduced to enable a new method of conducting above ground classifications in airborne LIDAR datasets beyond the standard classification routines. Prior to classification, the software groups above ground points with the goal of designating a single group of points per object. Each grouping of points receives a unique identifying number and this number is assigned to each point. The assigned group attribute may be saved in the FBI format for future use. The end game of these new grouping tools is to become the preferred method for classification of above ground features.
The usual method of classification contains some limitations that this grouping method improves upon. The older routines generally look at individual points to determine how each need to be classified. Some old classification routines form temporary groupings of points to assist in classification (i.e. planar points when doing building classification), but this grouping information is not stored. Each of these routines have their own grouping principle and there was no way to evaluate if a group is more like a tree or a building.
In short, the usual classification routines classify individual points, but the grouping method enables classification on an object basis. This approach in turn provides better automatic classification, while also fostering faster manual classification corrections of above ground features. The introduction of the classification by groups routines are by far the biggest improvement in the software in the past year.
Principles of Grouping
Grouping logic relies on information derived from the point cloud that determines how a point is related to adjacent points. The important information is in the form of the ‘normal vector’ and ‘distance’ attributes. Where the distance is the height of the point above the ground, while the Normal vectors are the local slope aspect of the data. From the normal vectors the dimensionality of point groups; linear, planar, or complex relationships with adjacent points can be ascertained.
Read Complete Article: TerraScan: Classify by Groups