Robust Confidence Measures for Multi-Temporal 3D Spatial Change Detection
Abstract
Research Problem: The identification of temporal differences in the three-dimensional (3D) structure of built or natural environments, i.e., spatial change detection, provides actionable information in the form of localized object detection and macro-scale landscape morphological changes. As 3D spatial remote sensing technologies such as lidar and dense image matching (a.k.a. structure from motion) continue to mature, an abundance of spatial data is available for change detection analysis. However, a fundamental deficiency exists in the majority of current spatial change products: an absence of statistically robust and spatially variable estimates of confidence in the reported change. This deficiency taxes the ability of analysts to discriminate signal and noise in spatial change products and precludes the ability of downstream algorithms to provide corresponding uncertainty estimates. The objective of the proposed research is therefore to increase the capability, accessibility, and exploitation potential of robust uncertainty estimates for geospatial change detection products through creation of an open source algorithm toolset that addresses current gaps in knowledge and practice. Technical Approach: Algorithm development will be centered on the existing open source Point Data Abstraction Library (PDAL). This will enable rapid and flexible development in a Python environment while leveraging PDAL’s robust 3D data ingest, manipulation, and export capabilities. Once proven effective, performance-sensitive algorithms will be ported to C++ if necessary. By employing PDAL and Python, future transition of the developed algorithms to applications such as GRiD (Geospatial Repository and Data Management) can be accomplished without special software development. Anticipated Outcome: In addition to several refereed journal publications, we anticipate generation of a set of algorithms that ingest raw lidar observations or point cloud data, produce rigorous lidar point uncertainties, render rasterized digital elevation model (DEM) uncertainty maps, and produce horizontal and vertical uncertainties in spatial change computed from temporally spaced DEMs. The toolset will be open source, licensed in such a way as to not impede commercial inclusion, and hosted in publicly available source repository to encourage implementation within the geospatial community. Impact on NGA Capabilities: This proposal directly addresses the interests of the NGA in advancing geolocation and data uncertainty in spatio-temporal analysis tasks. Specifically, the work in this proposal will provide methodology and algorithms necessary for generating confidence measures in spatial change derived from temporally spaced 3D point cloud products. Team: The proposal is led by Dr. Preston Hartzell (PI) with the University of Houston. The PI is supported by Dr. Craig Glennie (Co-PI, University of Houston) and will collaborate with Mr. Howard Butler of Hobu, Inc.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Oct 06, 2020
- Source ID
- HM04761912012
Entities
People
- Preston Hartzell
Organizations
- National Geospatial-Intelligence Agency
- University of Houston System