Develop an optical processing algorithm that employs multiple photographs taken at multiple known fixed (but separate) locations at the same time (or approximately the same time) to create a 3 dimensional model. This model may also include superimposing the images extracted from the photographs in order to recreate a much more realistic 3 dimensional model.
Processing will be based upon techniques similar to synthetic aperture radar- where a 1 dimensional radar return from a transmitter/receiver that moves along a fixed path is processed into a 2 dimensional image.
Identifying the "known path" is a vital part of this innovation. All position measurements are relative to some frame of reference, such as "my position and orientation at t=0" and each measurement techniques has its own limitations (inertial measurements drift, GPS degrades as the direct RF path to the satellites get blocked). By using target identification techniques and movement, position would be back calculated to improve the measurement. As the position fix improves, so should the errors of the 3 dimensional model decrease, and by iteration, the position errors would decrease as well- a limit that would ultimately be based upon the optical angular resolution of the cameras used. Ultimately, multiple methods should be used for position measurement, relying upon the relative strengths of each method.
This algorithm has many potential applications including:
Real estate: 3 dimensional walk-through models
Robotics: path planning and environmental model creation
Military/S.W.A.T.: Scenario creation for mission planning
To be more specific:
In a safety related robotics application, an assistant robot would follow a rescue worker into a building, recording images in many directions as it moves. These images are processed in real time to create the model that the robot could follow to retrieve an object for the rescue worker, or to lead the rescue worker back to a known point, and identify any changes since the initial pass through the building.
Urban planning/modeling would also be a potential application- a vehicle (or individual with a camcorder) would move along a path, and these images could be processed into a 3 dimensional model to be used for infrastructure surveys.