A Novel Building and Tree Detection Method From LiDAR Data and Aerial Images

In recent decades, building and tree detection from LiDAR data and aerial imagery with high automation and accuracy level has been the focus of many researchers which was selected as the purpose of our research. At first, after data preprocessing, off-terrain objects (OTO) including trees and buildings were extracted from LiDAR data. Second, a number of features were produced as inputs of support vector machines (SVMs) to separate buildings from trees. In the SVM, an automatic procedure was used for selecting the training data. After separating the trees, mathematical morphology operations were used to eliminate small objects and fill small holes in the detected buildings and trees. Finally, k-means clustering algorithm was used to separate buildings with different heights. The obtained results for detected buildings and trees were evaluated by working group III/4 of ISPRS, which demonstrated a high rate of success. For completeness, correctness, and quality metrics in per area mode, average values of 88.70%, 95.60%, and 85.30% for buildings and 74.30%, 63.50%, and 52.10% for trees were obtained, respectively.