The goal of this lab was to learn how to classify ground in LiDAR data. This would be done by learning the tools necessary to filter out noise points and classifying ground. Some basic QA/QC would be necessary to ensure accuracy.
METHODS
LP360 was used for this entire lab. First, the data was loaded into the program. Statistics were then taken through the Point Cloud Statistics Extractor, whose results are shown below.
The next task was to remove low noise points. These points were clearly shown using the profile view, which displayed isolated points far below the ground. First an automated point cloud task was run to filter out these noise points. After this was completed, the profile view was used to check points that the automatic filter could have missed. After the entire study area had been checked, I could move to classifying ground points.
First, seed points were needed to determine average ground levels. The point cloud task used to do this works by using windows of a given size to determine a single ground point on each window. The given size of these windows must be larger than the largest building in the study area so it does not choose a point on a roof. The size of the largest building was determined to be 500 feet. The point task was then executed, and created a grid of seed ground points.
The seed points needed to be check since not all landed on the actual ground. This was easily checked by having LP360 display only seed points using a TIN model. Any noise points chosen as seed points were shown as a sharp change of elevation, and could be reclassified. Once the seed points were determined to be accurate, the TIN model looked fairly flat, as shown below.
With accurate seed points chosen, ground points could now be classified. The same point cloud task was run as before, only this time to classify all ground points. After running and checking the results twice, it was determined good enough to work with for QA/QC.
To do QA/QC, I slowly scanned across the study area and fixed incorrect classifications from the algorithm. Most of these were points that were classified as ground when they were not, such as buildings.
After this was done, water was classified using a water break-lines shapefile.
The results are shown and discussed below.
RESULTS
The results of the data after the algorithms and manual cleanup are shown below. This is a view of the total study area after the beginning classification work was completed.
Note how the blue areas, water, have patches missing. This is due the absorptivity of water being high, leading low reflectivity and low signal return.
The results of the data after the algorithms and manual cleanup are shown below. This is a view of the total study area after the beginning classification work was completed.
Below are a couple of examples of closer views within the study area displaying how buildings are separated from ground. Orange is ground classification and grey is unclassified. The buildings will be classified in subsequent labs.
Note how the blue areas, water, have patches missing. This is due the absorptivity of water being high, leading low reflectivity and low signal return.
SOURCES
Data obtained from Cyril Wilson for use in 358 LiDAR course.
Data obtained from Cyril Wilson for use in 358 LiDAR course.
