Detection of 3-D Individual Trees in Urban Areas by Combining Airborne LiDAR Data and Imagery

Abstract

An automated approach to extracting 3-D individual trees in urban areas is developed based on jointly analyzing airborne LiDAR data and imagery. First, the spectral, geometric, and spatial context attributes are defined and integrated at the LiDAR point level. Then, a binary AdaBoost classifier is used to separate points belonging to trees from other urban objects. Once the classification is completed, a spectral clustering method by applying the normalized cuts to a graph structure of point clouds of the vegetation class is performed to segment single trees. The geometric and spectral attributes play an important role in establishing the weight matrix, which measures the similarity between every two graph nodes and determines the cut function. The performance of the approach is validated by real urban data sets, which were acquired over two European cities. The results show that 3-D individual trees can be detected with mean accuracy of up to 0.65 and 0.12 m for tree position and height. Based on the results of this work, geometric and biophysical properties of individual trees can be further retrieved.