Date of Award:


Document Type:


Degree Name:

Master of Science (MS)


Civil and Environmental Engineering

Committee Chair(s)

Robert T. Pack


Robert T. Pack


David G. Tarboton


James A. Bay


This thesis reviews algorithms that have been developed for classifying lidar data and identifies a progressive morphological filter for evaluation and improvement. Two potential weaknesses evaluated include the row-column order bias and grid orientation bias.

Four different row-column orderings were developed to test for bias associated with the order choice. Moreover, a method rotating the filter grid to a series of angles was developed for testing bias associated with grid orientation. Measures of success of the improvements include Type I and II errors, where results are compared with a hand-produced "truth" dataset. Two datasets, one urban, the other rural, were selected for testing the modified filters. The results are presented and discussed for each algorithm.

It was found that the four row-column orders all classified the dataset exactly the same. After the erosion and dilation functions were completed, the same surface profiles and elevations were produced regardless of row-column ordering. The filter windows used by the algorithm were found to create a rectangular area where the minimum and maximum values within that area were always selected. Therefore, it was found that the row-column orders did not create a bias in the classification.

However, grid orientation was found to greatly affect results. Misclassification problems occurred at ridgelines, mounds, and along roads with ditches and steep slopes running along them. Grid angles running parallel to these objects were found to avoid these errors. Buildings also created errors, but were minimized with grid angles crossing them at 45 degrees. The selected angle directions significantly affect the classification results in all cases. Therefore, the grid orientation bias was verified.

Two new methods of combining the results from the various angles have been developed. The first method used the best two classifying angles to combine the results. Best results were found in datasets with terrain objects positioned in similar directions for this method. The Multiple Angle method used all of the angle classifications to combine the results. This method performed best on datasets with terrain objects oriented in numerous directions. More accurate terrain models and better overall classification results have been generated using these methods.




This work made publicly available electronically on April 11, 2011.