3D Models from 2D Imagery: Raytheon’s Intersect Dimension

As you may know from previous columns, I’m a big proponent of oblique imagery and 3D models for non-GIS users such as first responders or the general public. The primary reason is that most non-geospatially-trained people have a hard time getting oriented when viewing only abstract maps or ortho imagery. Oblique views are the way we navigate our natural 3D world, so anything less is not ideal. 

Many products and services provide oblique imagery and 3D models. Pictometry, Bing (Pictometry imagery), and Google provide oblique views. Numerous companies build 3D models such as AEgis Technologies, Esri with City Engine, Cyber City 3D, and PLW Modelworks, which builds probably the most photorealistic and photo-accurate models.

Each company builds its 3D models or views with slightly different technology and methods. This year at GEOINT, Raytheon caught my attention with a 3D-model creation process that uses only 2D imagery, even satellite imagery of locations that are not accessible by ground or air. The system, called Intersect Dimension, doesn’t use LiDAR, but leverages passive 2D and other remote-sensing imagery sources to automatically create high-accuracy 3D models very quickly and for a fraction of the expense of current multi-dimensional modeling solutions. Dimension uses a Raytheon-patented technology to automate optimal image selection and geometric registration to build models with high positional accuracy and optimal resolution.

Unlike LiDAR-generated 3D data, Dimension’s passive 3D point clouds offer several advantages, including

• global coverage in airborne-denied geographies or in areas with outdated flyover information
• extensive coverage from multiple data sources
• compatibility with a broader selection of data sources, including archived (historic) images
• faster turn-around times with fully-automated renderings requiring no human intervention
• compatibility with existing investments in people, training and enterprise visualization/exploitation tools and systems.

Here is an example of a limited access location in Tripoli colored by elevation and alternately colored with true color RGB (see below).
Intersect Dimension, © Raytheon Company 2014,
Includes material © CNES 2014, Distribution Astrium Services / Spot Image Corporation, USA, all rights reserved,
Pleiades Source Imagery

The true color option of the above.
Intersect Dimension, © Raytheon Company 2014,
Includes material © CNES 2014, Distribution Astrium Services / Spot Image Corporation, USA, all rights reserved,
Pleiades Source Imagery

Dimension performs extraction of vertical feature points to build true-to-life models with walls that are perpendicular and details like tall, thin buildings, light poles and antennas accurately represented without tilting or a “melted chocolate” look.  Additionally, all extracted 3D features are geo-referenced with horizontal and vertical accuracy touted by Raytheon to be better than the original image sources. The image collections are processed using algorithms in a proprietary, patented photogrammetric bundle adjustment that increases the accuracy of XYZ locations.

Here is a screen capture of QT Modeler showing tower to tower measurements in front of the
Beijing “Bird’s Nest Stadium” that is colored by height.
Intersect Dimension, © Raytheon Company 2014, Includes copyrighted material of DigitalGlobe, Inc., Viewer courtesy of Applied Imagery

Multiple images are used to create the models. In fact, the more images the better for a final product. Metadata is maintained so users can understand the source and dates of the imagery used to build the models. The resultant models permit accurate measurements in three directions, XYZ, and angular measurements from one high point to another, such as towers or rooftops. Since the model is fully georeferenced, GIS vector data can be accurately overlaid on the model, and data layers such as flood planes can be displayed while showing where the plane intersects buildings.

The Dimension process is extremely fast. In a typical operational environment, Dimension can automatically produce a large-area photorealistic and photo-accurate model in minutes or hours, a process that typically could take weeks or even months using older technology.

The Intersect family of solutions enhances Dimension with data fusion, processing, analysis, visualization and automation capabilities using almost any data source or system, including full-motion video (FMV), activity-based intelligence (ABI), social-media tracking and multi-source data integration. Dimension allows for the addition of new analytics, rich content and augmented-reality data while enabling the fusion of data from public and private/proprietary data sources.

Here is a snapshot of a Turkish open-pit mine of a temporal 3D model built from multiple video images. The models are fully metric, permitting rapid change detection and measurement.
Intersect Dimension, © Raytheon Company 2014, Includes copyrighted material 2014 © Skybox Imaging, Inc., All Rights Reserved

A second shot of the same Turkish mine as above.
Intersect Dimension, © Raytheon Company 2014, Includes copyrighted material 2014 © Skybox Imaging, Inc., All Rights Reserved

Each point on the 3D model can be colorized with electro-optical data from source imagery, or points can be colored manually using external sources. Imagery can be used from any time of year or any time of day, is not constrained by collection angles, and is not limited to stereo pair collection.

Dimension models are compatible with existing enterprise investments in people, training, and visualization and exploitation tools. The models can be viewed, and interactively rotated, zoomed, measured and navigated, using many legacy 3D viewers such as Google, Skyline, and Autodesk. Occlusions or “no data” areas are left blank, but similar data can be cloned to fill in gaps if the customer desires the cloning. Trees are modeled similar to other types of reflective surface models. Since these are standard format digital models, a user can create wireframes or bare-earth models if desired.

You can contact the people at Raytheon for samples of even higher resolution models and additional information at analytics@raytheon.com. They also have video clips that demonstrate the system in operation so you can get a feeling of the system speed and operation in a more visually compelling way. The engineers at Raytheon have built quite a strong addition to our geospatial toolbox and they deserve your serious consideration.