Summary: | Forward Looking Infrared (FUR) systems are commonly used in military applications for the purposes of detecting and recognising moving or stationary objects. FUR imagery is particularly effective in low visibility environments and can provide additional information which would not be available in visible band images. The disadvantages of FUR imagery are that it tends to be extremely noisy, low contrast, and cluttered due to manufacturing limitations and environmental constraints. Contemporary research has mainly focused on applying detection and recognition techniques directly to FUR image sequences. However, compared with visible band images, FUR imagery has much poorer quality which results in greater difficulty in detecting and recognising objects. This thesis describes the development of techniques to improve the quality of FUR imagery prior to performing detection and recognition, with the aim of improving object detection and recognition performance. Super resolution and image mosaicing techniques have been employed for high-resolution assessment of individual areas and high-level situational awareness of large areas respectively. Both super. resolution and image mosaicing rely heavily on accurate image registration hence an image registration system with sub-pixel accuracy has been developed especially for FUR imagery. This image registration technique aligns imagery efficiently and accurately in spite of the inherent limitations of FUR images. Then, a robust and efficient super resolution method has been adopted to enhance the image resolution and a mosaicing method based on the super resolution method used to enlarge the field of view of the image. In addition, cloud effects have been considered and a segmentation scheme developed to deal with cloud cover on FUR imagery.
|