Unsupervised Clustering Method for Complexity Reduction of Airborne and Terrestrial 3D Point Cloud Data in Marshes

The resilience of marshes to a rising sea is dependent on their elevation response. Accurate characterization of a marsh environment is important for coastal management and conservation. This research proposes a novel unsupervised clustering method specifically developed for segmenting dense 3D point cloud data acquired over marshes by geodetic imaging techniques, such as lidar and structure-from-motion (SfM) photogrammetry. The work here explores segmentation of point cloud data acquired from the ground by active Terrestrial Laser Scanning (TLS) and from the air by passive RGB imaging using an Unmanned Aircraft System (UAS). SfM processing is applied to the UAS imagery to derive a 3D point cloud data of the marsh environment from an aerial “nadir looking” perspective. The fundamental idea behind this novel clustering framework is the application of voxel presentation of 3D space to create a set of features that characterizes the local complexity and geometry of the terrain. The combination of point and voxel features are utilized to segment 3D point clouds into homogenous group in order to study surface change and vegetation cover. Both TLS and UAS-SfM approaches can provide detailed topographic measurements of the terrain. The dense point cloud provides a 3D representation of the surface, which includes both terrain and non-terrain objects; notably vegetation cover. 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, such as a bare-earth digital elevation model. The results suggest that the combination of point and voxel features represent the datasets well. The proposed method is capable of identifying tidal flats, vegetation, non-natural features and in so doing, reducing the complexity of the scene. Within only eight clusters using the TLS data, it compresses a marsh surface into different clusters: tidal flat, mangrove, low marsh transition to high marsh, upland, and non-natural features. The method is also applied to cluster the UAS-SfM point cloud data acquired over the marsh. The framework is adapted to integrate the RGB reflectance information captured by the camera while replacing other TLS specific features to compare segmentation based on the UAS-SfM and TLS point clouds. Results show that the developed clustering framework is generalizable to 3D point cloud data acquired from different imaging modalities and can be applied to reduce data complexity within marshes for improved scene interpretation and change detection monitoring.