Search

The aim of this paper is to expose the significant contribution of colour information to the classification of 3D points from dense aerial image matching. In this study, colour information was used to compensate for the lack of additional information in photogrammetric point clouds with the proposed automatic hierarchical rule-based classification approach. Two point-based classification approaches based on only spatial-based features and combining the use of spatial-based features and colour information were developed in order to classify photogrammetric point cloud data. The hierarchical rules were developed using the selected features for the proposed point-based classification approaches. Detailed parameter analyses for the developed rules were carried out in pilot areas. The photogrammetric point clouds from the Zekeriyakoy test site in Istanbul were automatically classified, and ground, building, and vegetation classes were acquired using only spatial-based features and using both spatial-based features and colour information. The contribution of the colour information was clearly seen in the vegetation points classified as buildings in woodland and urban areas, as well as in the ground points classified as vegetation in urban areas. The performed accuracy analysis of photogrammetric 3D point cloud classification with the combined use of spatial-based features and colour information verified the improvement of overall accuracy from 70 to 78% in comparison to the classification of only spatial-based features. [ABSTRACT FROM AUTHOR]