Lab 1: LAS Ground and Water Classification

Goals and Background:

The main objective for this lab is to classify LAS preprocessed points into water and ground points which could then be used in many different applications. To do this the ground filtering algorithm is used to classify ground points and remove low outliers, manual cleanup is employed to complete classification, and water body feature breaklines are used to identify water points.

Methodology:

Part 1:

All LAS files from a flight over Lake County, Illinois were brought into LP360 64-bit for Windows with read and write abilities along with a raster NAIP image of the study area for contextual information. The data was then displayed by classification to see the initial state of the data: all points had not been classified (Figure 1).

Figure 1: All points unclassified

The live view window was then opened, and after filtering by all points the autoselect all classes being used on active LAS layer button was selected to show the classes the points displayed were classified as (Figure 2). This confirmed the earlier observation. Next, the points were classified by return combination which resulted in the view shown in Figure 3. It was observed that points in flat areas were mostly single returns, and that in wooded areas there were points classified as first and last of many points.

Figure 2: Use of the live view window

Figure 3: Return Combination View

Continuing through Part 1, statistics were generated per LAS tile, which were eventually stored in a shapefile that contained all LAS tile polygons. This included header information as well as point count, point density, number of flightlines, number of return numbers, area, and nominal point spacing (NPS), as well as minimums, maximums, and averages for intensity, return numbers, and scan angles. This was configured via adding a point cloud statistics extractor point cloud task and then configuring it (Figures 4 and 5). The statistics generated for each LAS tile could then be viewed by use of the identify tool.

Figure 4: Creating a new point cloud task

Figure 5: Configuring the new point cloud task

Part 2:

All processing done in Part 2 was conducted from a local drive for increased speed.

Section 1: Removal of low outlier points

A new task was created of the Low/Isolated Points Filter type. This was performed on only class 1 of points: unclassified. The low/noise class was selected for the low destination class. The rest of the properties of this task can be seen in Figure 6. These points were then observed using the live filter, force 100% resolution, and the profile tool. The profile tool revealed depth below the other points that the filtered points lied.

Figure 6: Removal of low points (noise)

Section 2: Automatic ground point filtering

Another task was added of the Adaptive TIN (triangular irregular network) Ground Filter type was added. This task was configured with the unclassified points as source points, ground classification as the destination class, and the seed sample distance as the distance across and along the largest continuous surface in the LAS tile coverage area. The rest of the parameters can be seen below in Figure 7.

Figure 7: Ground Classification

This last step resulted in seed ground points from which a tin could be created. The sample points needed to be culled first before they would be used. The process to do this required setting up a basic filter type point cloud task of subtype seed correction with the seeds that would be selected being changed to the classification desired. For example, Figure 8 shows points below the surface which should be correctly classified as Low Point (noise). Some points were also removed as ground seeds and reclassified as unclassified which were not ground but in fact roof.

Figure 8: Low points erroneously classified as ground seed points

The last step was a suggestion to play with reducing the seed sample distance so as to make sure that no areas would be under-seeded or to manually add seeds. Some under-seeding was observed in the resultant data from processing using the seeds that were previously generated and culled (Figure 10). For sake of speeding up the lab this step was made not required and so it was not performed.

Figure 10: Apparent under-seeding shown by arrow (north central)

Section 3: Qualitative accuracy assessment of automatic ground point classification

Using the 3D viewer and the profile window the automatic ground point classification was assessed. The results will be discussed in the results section.

Part 3: Manual cleanup of ground classification

Section 1: Correcting Omission Error

To correct for omissions of ground points the ground cleanup filter task was used with the parameters below. Polygons were then drawn around areas needing to be reseeded for ground classification, known ground points were selected to seed from, and the areas were reclassified.

Figure 11: Ground cleanup filter parameters

Section 2: Correcting Commission Error

The basic filter was utilized here similarly to part 2 in order to remove sections of points that were incorrectly classified as ground. The profile and 3D window were used to verify points were not actually ground.

Part 4: Classifying Water Features

Water features were classified by features in an imported shapefile and a simple classify by features point cloud task.

Figure 12: Classifying water features using a shapefile

Results:

The culmination of all of the above processes was a moderately accurately classified image in terms of objects, water features, and ground. Some issues did exist such as islands being classified as water, and buildings being classified as ground, but these were able to be fixed in manual cleanup.

Sources:

• LAS, tile index, and metadata for Lake County are from Illinois Geospatial Data ClearingHouse.
• NAIP imagery is from United States Department of Agriculture Geospatial Data Gateway. 
• Breaklines is from Chicago Metropolitan Agency for Planning. 
• Lab instruction was provided by Dr. Cyril Wilson.