The availability of high quality RGB images and LIDAR data provides efficient image
classification using the complementary properties of these data sources. The paper objective is to
introduce an automated urban unsupervised classification technique using combined semantic
(from RGB image) and spatial (from LIDAR data) information leading to the ability to extract
different features rabidly and efficiently. In this study, new concepts and techniques for mapping
urban areas using aerial images and LIDAR data fusion was developed and tested based on
available data. Genetic Algorithms (GAs) were integrated with different fitness functions to
produce two proposed techniques. K-Means (KM) and Fuzzy C-means (FCM) algorithms were
tested and compared, as fitness functions for GAs. Three groups of data were applied which
include: RGB group; RGB/ LIDAR data group and RGB/LIDAR/attributes group. Error matrix
and K-HAT (kappa) statistics were adopted as well as visual inspection to evaluate the
classification results. FCM proved to be a preferable fitness function for GAs-based classification
from aerial images and LIDAR data with accurate average classifications of 87.84%. The feature
classification techniques developed in this study are automated, efficient and present a suitable
method for extracting different features while overcoming most of the problems in situations of
similarity existing in texture or height information accompanied by fast and reliable results.
During the vectorization phase, the classified images were processed through a series of image
processing techniques to produce the digital vector buildings map. An accurate estimation of
buildings was carried out from the reference data and compared against the corresponding results.
The extracted building data coordinates were compared against the Global Positioning System
(GPS) observations and the standard deviation was 0.43m. |
