SBIR-STTR Award

Automated extraction of 3D objects from video sequences represents a key 21st Century enabling technology for DOD applications, ranging from interactive C3 systems to virtual reality mission rehearsal. While dozens of researchers have demonstrated underlying theory, significant development issues remain which have prevented introduction of any relevant systems. General Reality Corporation (GRC) proposes to attack this bottleneck by advancing the state-of-the-art in structure-from-motion (SFM) algorithms, through integration of edge extraction and aggregation for improved depth mapping, meshing, and object segmentation. Driven by a Phase II goal of automatically populating the New World Vistas Global Awareness Virtual Test Bed (NWV GA VTB) using video source data, GRC will establish Phase I feasibility by 1) developing a real-time capable, edge-assisted breadboard of its unique "scene to simulator" processing pipeline, 2) demonstrating its potential performance in a NWV GA VTB visualization, and 3) developing a complete system design for Phase II maturation and delivery. The proposed effort will utilize and extend GRC's strong technology base, which includes a DARPA-sponsored controlled imagery acquisition system and a Marine Corps-sponsored SFM architecture for semi-automatic video model generation.

Keywords:
3-D MODELING DEPTH MAPPING TERRAIN MAPPING SITE MODELING VIRTUAL REALITY VIDEO COMPRESSION GLOBAL AW

Automated extraction of 3D objects from video sequences represents a key 21st Century enabiing technology for DoD applications ranging from interactive C3 systems to virtual reality mission rehearsal. While dozens of researchers have demonstrated underlying theory, significant development issues remain which have prevented introduction of broadly useful systems. Geometrix proposes to attack this bottleneck by advancing the state-of-the-art in structure-from-motion (SFM) algorithms, through integration of edge extraction and processing for improved depth mapping, meshing, and object segmentation. The effort will develop sophisticated edge extraction and tracking algorithms, and use them to 1) constrain SFM depth mapping; 2) integrate structure-from-contour with SFM; 3) constrain meshing of depth maps to ensure that mesh lines match 3D edge locations; 4) segment objects of military interest from reconstructed scenes; and 5) match candidate objects against existing objects in a geospatial database. Following algorithm development, system-level solutions for handheld camcorder and hands-off aerial image acquisition/processing at low latencies will be integrated and delivered. The proposed effort will utilize and extend Geometrix's strong technology base, which includes a DARPA-sponsored controlled imagery acquisition system and a Marine Corps-sponsored SFM architecture for semi-automatic video model generation.

Benefits:
3D object recovery from video imagery is a ubiquitous problem within the $20B 3D graphics industry, where 250,000 3D content authoring professionals require dramatic efficiency improvements. Geometrix recently introduced the world's first commercial system for generating 3D models from video imagery, and is well positioned to rapidly introduce a new generation of products which take advantage of the Phase II results well before Phase II conclusion. Geometrix's current markets of movie special effects, game development, and on-line electronic commerce catalogs will be expanded to include military/flight simulation, urban planning, architectural engineering/surveying, and travel/real-estate marketing via the proposed effort.