Evaluation of the Self-Organizing Map Classifier for Building Detection from Lidar Data and Multispectral Aerial Images

Integration of aerial images and lidar data compensate for the individual weaknesses of each data set when used alone, thus providing more accurate classification of terrain cover, such as buildings, roads and green areas, and advancing the potential for automation of large scale digital mapping and GIS database compilation. This paper presents work on the development of automatic feature extraction from multispectral aerial images and lidar data. A total of 22 feature attributes have been generated from the aerial image and the lidar data which contribute to the detection of the features. The attributes include those derived from the Grey Level Co-occurrence Matrix (GLCM), Normalized Difference Vegetation Indices (NDVI), and standard deviation of elevations and slope. A Self-Organizing Map (SOM) was used for fusing the aerial image, lidar data and the generated attributes for building detection. The classified images were then processed through a series of image processing techniques to separate the detected buildings. Results show that the proposed method can extract buildings accurately. Compared with a building reference map, 95.5 percent of the buildings were detected with a completeness and correctness of 83 percent and 80 percent respectively for buildings around 100m(2) in area; these measures increased to 96 percent and 99 percent respectively for buildings around 1100m(2) in area. Further, the contributions of lidar and the individual attributes to the quality of the classification results were evaluated.