A Hierarchical Unsupervised Method for Tree Detection Using Aerial Imagery and LiDAR

Tree detection using aerial sensors in early decades was focused by many researchers in different fields including Remote Sensing and Photogrammetry. This paper is intended to detect trees in complex city areas using aerial imagery and laser scanning data. The methodology used in this paper is a hierarchal unsupervised method consists of some primitive operations. This method could be divided into three sections، in which، first section uses areal imagery and both second and third sections use laser scanners data. In the first section a vegetation cover mask is created in both sunny and shadowed areas. In sunny areas Normalized Difference Vegetation Index (NDVI) and in shadowed areas a Shadow Index (SI) is employed to obtain vegetation cover mask. This mask contains grasses and bushes that should be eliminated in the other two sections. In the second section Rate of Slope Change (RSC) is used to eliminate grasses. Such areas that their RSC is below a threshold is eliminated. In the third section a Digital Terrain Model (DTM) is obtained from LIDAR data. By using DTM and Digital Surface Model (DSM) we would get to Normalized Digital Surface Model (nDSM). Then objects which are lower than a specific height are eliminated. Now there are three result layers from three sections. At the end multiplication operation is used to get final result layer. This layer will be smoothed by morphological operations. Our result has a good rank in comparing to other methods in ISPRS WG III/4، when assessed in terms of 5 indices including area base completeness، area base correctness، object base completeness، object base correctness and boundary RMS. With regarding of being unsupervised and automatic، this method is improvable and could be integrate with other methods to get best results.