Automated 3D Reconstruction of a Scene From Persistent Aerial Reconnaissance Video at High Zoom

ABSTRACT: 3D scene model generation is a valuable tool for mission planning, precision targeting and sensor data exploitation.Our team of Systems & Technology Research (STR) and Vision Systems Inc. (VSI) proposes to develop automated 3D reconstruction techniques from narrow field-of-view video imagery.Such imagery presents multiple challenges:limited ground coverage, variable image resolution, and limited viewpoint diversity.STR and VSI have developed a voxel-based reconstruction framework called Belief Propagation Volumetric Reconstruction (BPVR) which excels at reconstructing detailed 3D models from aerial imagery.This technique simultaneously optimizes constraints from all data sources to produce a single composite 3D model, and works particularly well in challenging textureless areas.However, one current weakness of this algorithm is its dependency on viewpoint-rich orbitsthe 3D model quality degrades with low viewpoint diversity.Stereo-based 3D reconstruction approaches, on the other hand, perform well with limited viewpoints.The downside to stereo approaches, such as Semi-Global Matching (SGM), is that they output depth maps for each image and not a single 3D world.We propose to combine the performance of stereo under low-viewpoint diversity with the principled integration of the BPVR framework to generate an accurate and robust composite 3D scene model from narrow field-of-view video.; BENEFIT: The goal of this SBIR program is to provide users, such as Special Forces analysts, with high-precision 3D models for mission planning purposes by providing them with automated 3D reconstruction tools that replace their current manual model building pipeline.Our 3D reconstruction capabilities will be designed and validated to process practical sources of scene imagery such as narrow field-of-view video from unmanned aerial vehicle platforms.In particular, our tool will be robust to limited coverage area, variable image resolution, limited viewpoint diversity and variable image quality.In addition to developing a valuable tool for DoD image analysts, the robust 3D modeling capability has commercial applicability for drone security and geospatial data analysis problems.Commercial applications thus include drone system plug-in modules as well as geographic information system tools.