SharpGIS

#GIS from a .NET developer's perspective

Area classification based on laserscanned data

Midtvejs- and specialproject at IMM, DTU - January 2002
Download full report (danish)

Abstract (danish version below)

This report has been written on the basis of a project dealing with the practical application of laser-scanned data for area classification. The project is based on data scanned in the Spring of 2001 in Djursland on the Danish mainland of Jutland.

The report firstly introduces the concept of laser-scanning and the data available is described. The most significant sources of potential error in conjunction with laser scanning are then discussed and filtering techniques are applied in an attempt to eliminate such errors.

Thereafter, analysis and interpretation of various direct and derived parameters extracted from the collected data are made. This is done both graphically and using histograms. Statistical characteristics are sought which enable identification of various area classes from one another. Finally an attempt is made to construct an automated classification using a Bayes classification, and the result is evaluated.

It is concluded that is possible to differentiate between several types of area classifications on the basis of laser-scanned data. It is shown that the Bayes method can be used to differentiate between open country and wooded areas with a reasonable degree of certainty. However, if a more detailed classification is required, other methods should be considered.

Synopsis

Denne rapport er blevet til på baggrund af et projekt der omhandler laserskanningsdata og muligheden for anvende disse til arealklassifikation. Projektet tager sit udgangspunkt i data skannet i foråret 2001 over Djursland.

Rapporten indledes med en kort general introduktion om laserskanning og de data som er stillet til rådighed beskrives. De vigtigste fejlkilder i forbindelse med laserskanning gennemgås og data forsøges filtreret for fejl. Herefter analyseres og tolkes forskellige direkte og afledte parametre i den indsamlede data. Dette gøres både visuelt og ved hjælp af histogrammer. Der søges efter statistiske karakteristika der kan adskille forskellige arealklasser fra hinanden. Til sidst forsøges det at opstille en automatiseret klassifikation ved hjælp af en Bayes klassifikation og resultatet vurderes.

Det konkluderes at det er muligt at skelne mellem flere typer arealklasser ud fra laserskanningsdata. Det bliver vist at Bayes-metoden er i stand til at skelne mellem mark- og skovområder med en rimelig sikkerhed. Ønsker man en bedre eller mere detaljeret klassifikation, bør man dog overveje en anden metode.

Arcview PlugIn

During the project an ArcView Plugin for processing of laserskanned (lidar) data was developed. http://www.student.dtu.dk/~s973466/lasertool/

Illustrations

Z values - first signal
Z values - last signal
dZ values (Heightdifference in the two signals)
Intensity values - first signal
Intensity values - last signal
dI values (Intensitydifference in the two signals)
Orthophoto
Topographical map