PROCEEDINGS ARTICLE | April 29, 2009
Proc. SPIE. 7307, Airborne Intelligence, Surveillance, Reconnaissance (ISR) Systems and Applications VI
KEYWORDS: Cameras, Sensors, Image segmentation, Image processing, Video, 3D modeling, Image registration, Navigation systems, Algorithm development, 3D image processing
The overall goal of the research project reported here is to create a novel system that can combine input from multiple
passive sensors at different viewpoints (such as uninhabited aerial vehicles) into a single integrated three-dimensional
(3D) view of a scene. This form of intelligent data processing, known as Volume Registration, can further exploit the
available information to enable improved surveillance, reconnaissance and situational awareness, and thus offers
substantial potential benefit to military applications. This paper focuses on the case of multiple sensors onboard UAVs
operating at mid-altitude, and describes two complementary techniques that have been investigated in parallel to address
this challenge. The first of these is depth from disparity, which allows a real-time per-pixel estimation of the distance of
scene objects from the camera; the second is shape from silhouette, which back-projects a segmented version of the
image onto a 3D block of voxels and 'carves' a 3D model over multiple frames. The main steps of each algorithm are
outlined, along with appropriate results, in order to demonstrate how they could form a useful part of a practical Volume
Registration system. A number of possible extensions and improvements to the system architecture are also discussed to
improve the accuracy and efficiency of these techniques, and their applicability to the more complex low-altitude case is
discussed.
