3D Point Cloud Registration and Change Detection for Large-Scale Topographic Assessment

As LIDAR (light detection and ranging) sensors become cheaper and more accessible than ever before, 3D reality capture is advancing the field of computer vision in a way that provides significantly more intelligent insights than 2D photos and videos alone can. One of the cutting-edge applications of computer vision is point cloud change detection. Fundamentally, change detection allows one to observe temporal differences in terrain, objects, buildings, or vegetation over different timeframes. For example, change detection applied to a pre- and post-hurricane event captured by LIDAR point clouds can quickly identify damages to roads, buildings, erosion volume, flood paths, and other areas of interest. This is a powerful tool that can enhance situational awareness both on the battlefield and for preventative maintenance. In general, 3D point cloud change detection has been mostly relegated to manual point cloud registrations (the alignment of two or more different point cloud files) and geometric classifications. However, emerging machine learning techniques are pushing the boundaries of automated 3D algorithmic generation; it is now possible to quickly align and compare large-scale point clouds that would otherwise be both time-consuming and labor-intensive to produce. Stitch3D is developing and testing three separate deep learning models that algorithmically registers and calculates the changes between two or more large-scale terrain point clouds. Specifically, Stitch3D and our Research Institute partners at Oregon State University will be testing the k-nearest neighbors, naďve Bayes, and support vector machines (polynomial kernel function) approaches to determine the most accurate and efficient methodology to automatically align and identify differences between two or more point clouds. If successful, research resulting from this STTR Phase One effort will allow civil engineers and airfield managers to rapidly identify terrain changes – whether from hostile activity, natural disaster, or wear and tear – to support expeditionary missions and preventative maintenance requirements.