High Resolution Infrared Mosaicing Using Geo-Referenced Image
Valérie Lavigne
André Zaccarin (Supervisor)
Problem: Search and rescue operations are often carried out at night or under other conditions of poor visibility such as smoke, fog or snow. This is why military surveillance and detection systems are generally equipped with infrared detectors. Detection requires a wide field of view whereas identification requires a high resolution. This presents a great challenge because these concepts are opposed and it is generally necessary to carry out a compromise between the two.
Motivation: The Infrared Eye project was developed at DRDC-Valcartier to improve the efficiency of airborne search and rescue operations. A high performance opto-mechanical pointing system was developed to allow fast positioning of a narrow field of view with high resolution, used for search and detection, over a wide field of view of lower resolution that optimizes area coverage. This system also enables the use of a step-stare technique, which rapidly builds a large area coverage image mosaic by step-staring a narrow field camera and properly tiling the resulting images. The resulting image mosaic covers the wide field of the current Infrared Eye, but with the high resolution of the narrow field. This technique could eventually lead to a lighter and more compact system with the elimination of the wide field camera of the IR Eye.
Approach: First, a model of the complete system was created. Using this model we have build a flight image taking simulator that was used for the strategy conception and testing. A dynamic strategy that adapts itself to the operational context was favoured. It selects the next image position to minimize the presence of holes and the prisms displacement and to maximize the refresh rate. Image positioning in the mosaic is achieved using projection coordinates and homography methods.
Challenges: When we are in an airborne platform, the point of view of the camera changes constantly and images that were taken previously must be corrected to reflect the current point of view. Since the narrow field can be positioned anywhere within the wide field coverage, we need an acquisition strategy to decide where to take the next image.
Applications: This step-stare technique will be used to improve airborne search and rescue missions. The dynamic strategy developed is general enough to be used with other flight systems in any wavelength if corresponding Risley prisms can be found and flight data are provided to the algorithm.
Expected results: The step-stare technique brings the detection capability equivalent to that of the narrow field in the whole large area image. A dynamic strategy that enables the mosaic creation for display on the operator's console and controls the prisms to select the images to take was developed. A numeric simulator was also created and used to conceive, optimize and test the mosaic building algorithms.
Calendar: The strategy developed will be the object of a presentation at the SPIE Defense and Security Symposium in Orlando, Florida taking palce on April 12 - 16, 2004. The next step will be to test the strategy with a real system. One important factor to evaluate is in what extent do errors of measure affect the display. If errors are too visually unpleasing and distractive, correction methods could be added to the algorithm. Realtime experimentation is planned using a camera and a Risley prism set from a previous version of the IR Eye, and a rail with carriage mounted on the roof of a building to simulate scaled down flight displacements.
Support: This project is developed in collaboration with Defence R&D Canada - Valcartier. Financing is also provided through a FQRNT scolarship.
Last modification: 2007/09/28 by vlavigne


©2002-. Computer Vision and Systems Laboratory. All rights reserved