Chuyen T Nguyen1, Michael J Starek1,2, Philippe Tissot2
1College of Science and Engineering, Texas A&M University-Corpus Christi USA
2Conrad Blucher Institute for Surveying and Science, Texas A&M University-Corpus Christi USA
The resilience of marshes to a rising sea is dependent on their elevation response. Terrestrial laser scanning (TLS), commonly known as terrestrial lidar, is a detailed topographic approach for accurate, dense surface measurement with high potential for monitoring of marsh surface elevation response. The dense point cloud provides a 3D representation of the surface, which includes both terrain and non-terrain objects. Extraction of ground information in a marsh environment is a difficult task because of the occlusion of the laser pulse by varied types of dense vegetation. The extraction of topographic information requires filtering of the data into like-groups or classes, therefore, methods must be incorporated to identify structure in the data prior to creation of an end product. A voxel representation of three-dimensional space provides quantitative visualization and analysis for pattern recognition. The objectives of this study are threefold: 1) apply a multi-scale voxel approach to effectively extract features from the TLS point cloud data, 2) investigate the utility of K-means and Self Organizing Map (SOM) clustering algorithms for the segmentation of the data based solely on point cloud geometry and texture features, and 3) utilize a variety of validity indices to measure the quality of the result based on the compactness and the separability of the resulting cluster.
TLS data were collected at a marsh site along the central Texas Gulf Coast using a Riegl VZ 400 TLS. The site consists of both exposed and vegetated surface regions. To characterize the 3D structure of the TLS point cloud data, octree segmentation is applied to create a tree data structure of voxels containing the points. The algorithm works first by creating voxels based on the maximum number of points that a voxel can contain. If more points exist, the voxel will be recursively subdivided. Each voxel may be recursively decomposed into eight child voxels. The number of points in each voxel is varied as is the size. This adaptable structure captures characteristics of the TLS scan; farther away from the scan, point density decreases and the voxel adapts. The flexibility of voxels in size and point density makes this algorithm a promising candidate to locally characterize the geometry of the terrain. For example, the multi-scale voxelization can effectively characterize different types of vegetation because a small capacity voxel cannot store enough points to accurately extract features at a low density area, and a large capacity voxel cannot capture more detailed object shapes. Once voxelization is applied, statistical features of the point cloud belonging to each voxel are computed including the mean, minimum, and standard deviation of the point height, and geometric features are computed including the normal vector and curvature of the surface. The characteristics of the voxel itself such as the volume and point density are also computed and assigned to each point as are laser pulse interaction characteristics including deviation of pulse width. The features extracted from the voxelization are then used as input for clustering of the points using the K-means and SOM clustering algorithms. Four clustering evaluation criterions are assessed to determine the optimal number of clusters: Calinski Harabasz (variance ratio criterion), Davies Bouldin, Gap, and Silhouette. Results for different combinations of the feature space vector and differences between K-means and SOM clustering will be presented. The developed method provides a novel approach for multi-scale segmentation of TLS point cloud data in marshes that can be applied to better characterize the terrain, such as for vegetation biomass studies or erosion monitoring.