¶ Classification Guides
In this article, we explain the classification methods of the PCS software for semi-automatic and manual approaches.
¶ Semi-Automatic Classification
The PCS software is designed to support the operators with tools that make the extraction more convenient. It is not a purpose of the software to fully automate processes, as a fully automated task shall be crosschecked anyway by an operator. From our developer's perspective, it is a better way to provide semi-automatic tools that help the user perform the tasks more quickly and efficiently. The classification tools are developed with this mindset.
Before reading this article, please make sure you are familiar with the Classification concepts reading the respective article. This article only refers partly to vectors. Reading the Basic or Advanced extraction guide is not mandatory, but a general understanding of PCS's relation to vectors is recommended.
IMPORTANT! - The classification can be initiated only on LAS files; all points are loaded to the memory, as described in the optimizing article. A minimum of one LAS file shall have all points in memory; otherwise, the classification toolbars remain inactive. Also, it is impossible to load more GB of point clouds with all points in memory to the software than physical RAM installed on the PC.
In this guide, we provide the user with a few semi-automated scenarios for common use cases, but it's up to the user to get the most out of the provided tools.
The presented solutions are classified by LAS parameters and ground classification based on approximate TIN surface. It is also recommended that the tools available in the main window's classification toolbar and 3D view classification toolbar be used.
¶ LAS parameter classification
For this example, a single LAS tile has been used. The parameterized solutions can be applied to the whole or the SHP area (practically a polygon SHP). These tools are available on the main window's classification toolbar under the Class dropdown menu.
By starting the Whole area or Shape area command, the classification settings panel will be prompted.
The user can select the source classes (multiple) and the target class (single) in this setting panel. The classification will be affected based on this setting. It is not mandatory to use all classes; specific classes can be used as well, which can be helpful to reclassify a whole class to something else or reclassify within a polygon for all classes into a single target class.
There are six filters, where the user can specify ranges by some parameters to filter the classification results. If no further parameters are selected, the whole area/shape area will be reclassified, and no other parameters will be considered. The reclassification might take a longer time.
¶ Ground classification using approximate TIN surface
This method requires the user to be familiar with surface generation or have an existing surface in TRI format. This method utilizes the Lattice Export function to have a reduced-density point cloud, based on which a rough ground surface can be generated. The by distance from the TIN re-classification tool can be used from the main window's classification toolbar at the Class dropdown menu. The function can be used with surfaces generated from vectors only as well, but assuming the ground class is required for the vector extraction, the lattice method might be more convenient.
For the current example, the advanced extraction guide's sample was used, and the surface was built only from the vectors.
The TRI file shall be the active element when the tool starts. After the tool has been started, the settings panel will appear:
In this panel, the respective TRI file has already been pre-selected. The user can adjust the source (multiple) and target (single) classes. The details button lets the user filter the target classes by the classification settings filter. By default, it shall be empty; if a filter is applied, the window will show the active filter. The area selection can be measured as a polygon or a whole area. From the aspect of this tool, the entire area considers the whole TIN area, not the whole point cloud. The measured polygon allows the user to draw a polygon over the surface in 2D view, and the fenced area will be the target for the re-classification. The Max and Min height sets the distance above and below the surface to be reclassified to the target class based on the set parameters. After pressing OK, the tool will reclassify, which might take some time. If the tool overloads the classification buffer, the software will prompt the user to clear or increase the buffer size. Please read the classification article to understand the classification buffer better.
As mentioned above, the lattice method can also quickly develop a ground class in any cloud before vector extraction, which can help extract dense environments.
¶ Manual Classification Guide
The manual classification processes in PCS can be used to correct classification errors or convert a dataset for different types of use, for example, creating a sample dataset for machine learning scenarios, where a specific kind of object shall be trained for an ML engine from the point clouds.
Before reading this article, please make sure you are familiar with the Classification concepts reading the respective article. It is highly recommended that you get familiar with the clip frame toolbar and its mechanics. This article does not refer to vectors. It is not mandatory to read the Basic or Advanced extraction guide.
IMPORTANT! - The classification can be initiated only on LAS files; all points are loaded to the memory, as described in the optimizing article. A minimum of one LAS file shall have all points in memory; otherwise, the classification toolbars remain inactive. Also, it is impossible to load more GB of point clouds with all points in memory to the software than physical RAM installed on the PC.
The manual tools in PCS are based on a planar approach, which means that only planar views are sufficient for reclassifying something with the manual method. This means the 2D or the 3D view using the profile mode from the Clip Frame toolbar. PCS offers two primary types of manual reclassification tools: the polygon toolkit and the brush.
As the name suggests, the polygon toolkit can fence polygons in 2D or 3D view - when the profile mode is active - and the fenced area, or below or above the selected line, will be reclassified. This rougher tool allows the user to reclassify bigger chunks of the point cloud. Polygon toolkit can be found in the Main Classification toolbar and the 3D view Classification toolbar.
Using the 3D view classification toolbar, as with the polygon tool in 2D view, will reclassify everything in the marked area and not consider the elevation, except the classification settings have not been set specially for this.
The other tool is the brush. It also requires the profile mode for the 3D view. The brush does allow the user to perform a fine classification by hand. After the tool has been started, the user can set the source-target classes, set the pixel size, and turn on track mode (which is recommended). This will allow the user to draw - like using the brush in paint - and reclassify for the “painted” area.
The best practice for classification tasks for manual processes is to crosscheck the object from 2 sides, for example, front and left. Swapping the sides of the clip-frame tool from the Clip Frame toolbar can be handy for this task, as the perpendicular crosscheck can eliminate most of the manual classification errors.
After the classification has been performed, always save the modified point cloud. If it is not saved, the save warning will also prompt the LAS files for saving, but right-clicking on the Clouds inside the Project Explorer and selecting save is also a good approach.