Lab 1: LAS Ground & Water Classifcation

Goals and Background

The primary purpose of the following lab is to provide hands-on practice classifying LAS preprocessed points. Classification of ground and water points will be the main focus in the first lab. Classification in the most difficult and time demanding process when processing lidar data. During the lab I will be exploring how to utilize automatic ground filtering algorithms, the removal of low outlier points, manual cleanup of classification points, and classifying water utilizing breakline shapefiles.

Methods

Before starting any of the work a copy of the lidar was made to not corrupt the original data. 

All of the following work was performed in LP360.

Before beginning any of the processes I imported the data which included the LAS data and a NAIP image of the study area which was provided to me by my professor (Fig. 1). I inspected the data by displaying it in various options such as elevation, classification, intensity.

(Fig. 1) LAS data and NAIP imagery opened in LP360.

I then utilized Point Cloud Statistics Extractor found under the Point Cloud Task to calculate statistics on the LAS data. Under the General tab I set the parameters to extract the following: 

• Point Count
• Point Density 
• Number of Flight Lines
• Number of Return Numbers
• Area
• Nominal Point Spacing

Under the Point Attributes tab I set parameters to extract the following:

• Intensity (Min, Max, Average)
• Return Number (Min, Max, Average)
• Scan Angle (Min, Max, Average)

Ground Point Classification

Removal of low outlier points

The first step to acheiving an accurate classifcation of ground points is to remove low outlier points. I utilized the Low/Isolated Points Filter in the Point Cloud Tasks. The parameters were set per my professors instructions (Fig. 2).

(Fig. 2) Low/Isolated Point Filter parameters as displayed in the Point Cloud Task tab.

The filter doesn't remove all of the low outlier points so some manual cleanup is necessary after the next step to achieve high accuracy in the final ground classification (Fig. 3).

(Fig. 3) Low outlier points are displayed in purple. You can see gray points (unclassified) intermixed with the purple. The gray intermixed points have not been classified as low outlier points as they should have been.

Automatic ground point filtering

The second step in the classification of ground points required me to utilize the Adaptive Tin Ground Filter found in the Point Cloud Tasks to calculate Seed Points. Seed points are the lowest elevation found with in each grid square and are the basis of the TIN model. The parameters were set per my professors direction (Fig. 3). The source points were set to Unclassified since none of the points were classified at this point of the lab. The Destination Class was set to Ground. All of the Flag settings were set to Ignore for the purposes of this exercise. To set the Seed Sample Distance I opened the NAIP imagery in ArcMap and measured the longest dimension of the larges building within the study area. I determined the Seed Sample Distance should be set to 500 feet.

(Fig. 3) Parameters set to calculate Seed Points in the Point Cloud Tasks tab.

With the Seed Points created I now had to go through and preform manual cleanup of the Seed Points which were created on low outlier points. I utilized the profile view to inspect each Seed Point to determine if it was a true Ground Point. The majority of the Seed Points were classified correctly but a few throughout the entire study area were actually below the ground surface which is a result of low outlier points not being classified properly. I created a Basic Filter from the Point Cloud Task menu to change the classification from a Ground point back to Unclassified (Fig. 4).

(Fig. 4) Seed Point Cleanup filter parameters set in the Basic Filter option of LP360.

After the cleanup was completed I adjusted the settings of the Adaptive Tin Ground Filter to calculate the ground points (Fig. 5). After the filter processed the ground classification some of the areas which should have been classified as ground were left unclassified. I reran the Adaptive Tin Ground Filter to help fill in the void areas.

(Fig. 5) Zoomed in view of the Ground classification after running the Adaptive Tin Ground Filter.

After rerunning the filter there were still void areas which should have been classified as ground (Fig. 6). I preformed a manual cleanup of the ground classification using the Ground Cleanup Filter found in the Point Cloud Task tab. I utilized the Execute Point Cloud Task by Polygon to carefully outline areas which were ground but left unclassified. Once the tool was executed the unclassified points where changed to ground for those points which fit the parameters to ground (Fig. 7). Additionally I utilized the basic filter which I used to remove the erroneous seed points above to remove points which were classified as ground but were actually elevated such as the tops of buildings.

(Fig. 6) Ground area left unclassified by Adaptive Tin Ground Filter.

(Fig. 7)

Qualitative accuracy assessment of automatic ground point classification

I utilized the Profile Window and the 3D Window to inspect the classification. The shortage of Seed Points through the data caused a number of errors in the classification (Fig. 8). Highly accurate classification would have required a higher number of Seed Points then were created by the algorithm. Area with dense vegetation improperly classified as ground along with a few roof tops of buildings.

(Fig. 8) Profile view window and 3D view open to perform QA/QC on the ground classification.

Water Classification

I had to classify the water areas of the lidar data for the last portion of the lab. I employed the basic method of Classify by Feature using already created Breakline shapefile (Fig. 10). I created a new task with the Classify by Feature filter in the Point Cloud Task tab (Fig. 9). Under File Geometry File I set the Type to SHP Layer and the Map Layer to the Lake Breaklines file I was provided. The Source Points were set to Unclassified and Ground. The spatial relationship was set to Completely Within. The Destination Class was set to Water.

(Fig.9) Water Classification parameters set in the Point Cloud Task Window.

(Fig. 10) Breakline shape file open in LP360.

(Fig. 11) Sample of a water area classified by the algorithm in LP360.

Results

The results are far from perfect. There is always some error in LiDAR classification. Numerous days or weeks could be consumed to clean up all of the erroneous points.

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.