An Improved Classification Approach for Lidar Point Clouds on Texas Coastal Areas

There are two challenges in classifying lidar points into ground and non-ground points on Texas coastal areas, which usually has a low-lying landform consisting of morphological features including dunes, tidal and river channels with levees, barren flats, buildings, and trees with varying cover density. The first is to remove buildings and trees meanwhile keeping seawall, dunes, levees and channels. The second is to remove bushes and grasses. In this paper, a novel classification approach based on slope and neighbor properties is designed to meet these challenges. The innovation of this approach is to first determine the most suitable post-spacing for a given lidar point dataset and then to generate a raster with the post-spacing. Slope thresholds for landscape objects, such as buildings and trees, are derived from their own characteristic size. The classification has three main steps. Step 1 – identifying potential areas by removing steep slope cells. The slope calculation and removal are repeated several times. This step may incorrectly create holes. Step 2– restoring holes: the lidar points falling in a potential area are identified into two classes: the correctly removed or not. The latter are restored. Step 3– identifying bushes and grasses based on slope. Classifications have been carried out with a lidar point dataset of Mustang Island, Texas (a 40-km long barrier island) with promising results.