Lab 2: LAS Building Classification

Goals and Background: 

This lab practiced the classification of buildings in the same LAS point cloud dataset where ground was previously classified in Lab 1. The planar filter was used to classify the building points in LP360. After this filter was run, manual cleanup was performed on the dataset.

Methodology:

Part 1: Execution of the planar point filter

The planar point filter was used to classify buildings using the macro task type in the point cloud task window. A new macro task was made and then a planar point filter subtask. The unclassified class was used for source points, the ground class for ground points, and the building class for destination class. The maximum height parameter was calculated using the elevation of the tallest building with the lowest elevation found that was still classified as ground subtracted. This made sure that no taller vegetation was classified as buildings. The rest of the parameters were obtained form recommendations in the assignment instructions. The minimum height filter made sure that small planar objects were not classified as buildings. The minimum plane edge parameter also functioned to make sure that no small planar objects were classified as buildings and ensured that small objects that were not planar weren't picked up. The N Threshold set the maximum orthogonal distance to a plane a point could be and still be classified as a point. The plane fit adjusted the tightness of fit necessary for a planar surface to be defined and classified. Minimum and maximum slopes were set to limit the types of surfaces that could be detected as roofs. The maximum grow window area was set to the largest area of a roof in the dataset.

Table 1

The planar point filter was first run in a small section of the dataset with the execute by polygon tool to test the parameters, and after verification was run for the entire dataset.

Part 2: Manual cleanup of the classification algorithm results

Commission error was solved with basic filter type task setup with the source points set to building and the to classification set to unclassified. Using the 2D profile and 3D windows area were inspected for vegetation and other points that were wrongly classified as buildings. After commission error was fixed, omission error removal was performed with a separate basic filter task.

Results:

This lab resulted in LAS point cloud data with ground and building classified. Many errors in both classifications exist in certain errors as there is only so much a training eye and an algorithm can find, but many cleanly classified areas exist as well.

Sources:

• LAS Dataset for Lake County is from Illinois Geospatial Data ClearingHouse.
• NAIP imagery is from United States Department of Agriculture Geospatial Data Gateway.
• Breakline data is from Chicago Metropolitan Agency for Planning.