A Semi-Automatic Approach for Roof-Top Extraction and Classification from Airborne LiDAR

Airborne LiDAR provides us point cloud of the topographic features of an area. Point cloud classification is important to recognize which points corresponds to which target. Researches has been carried out for the extraction of building, trees, electric lines. But only few researches have been carried out for classification of different types of roof like flat, inclined and dome shaped. This research is aimed to achieve a semi-automatic approach to classify buildings and further classify the roof top type into flat, or inclined. Four subsets were taken from the LiDAR dataset, depending on the roof type. Initially, all the ground points are removed and non-ground points are segmented out. Later, the roof points of the buildings are classified on the basis of inclination into flat, inclined or dome type roof. A tool was generated in the Arc scene software using model builder. In which the subsets were used as the input and the different types of roof were classified. The accuracy assessment was done to calculate how accurately the classified points obtained belongs to the flat, inclined or dome roof tops. For all the four subsets, the overall accuracy for the flat, inclined and dome type roof obtained were 78.26%, 89.62% and 72.94%. This semi-automatic approach for the roof top classification is limited to categorize into flat, inclined or dome roof top only. Further, this research can be extended for the automatic classification of roof types and increase the accuracy.