It has long been known that (aided) target recognition based on imaging sensors is fundamentally limited by the number of pixels on the target imaged. Very roughly, at least 10 pixels are needed for detection, at least 100 are needed for generic classification, and perhaps 1000 are needed for recognition. While the details of this structure depend on specific sensors, targets, and sensing conditions, this general trend is universally observed in all domains of ATR. Doctrinal requirements of Total Information Dominance thus lead to massive deployment of sensors. How to gather, store, communicate, fuse, and analyze that vast flood of data in (near) real-time is then a formidable technical challenge. Moreover, given the lengthy DoD acquisition cycle, the fielded sensors are always inadequate the warfighter needs yet more pixels, to see more clearly. Without improving the sensors, there still exists the opportunity to improve their imaging resolution in "software" e.g., post imaging. The above analysis then points to the dramatic impact that post-imaging resolution enhancement, if available, can have on target recognition downstream. Video image resolution enhancement is the science/art of taking multiple frames from imaging sensor(s) that cover overlapping (but not necessarily identical) scene content, and integrating or merging the content for improved resolution for scene analysis, surveillance, situation awareness, ATR, and other applications. Fusing images for enhancement from the same sensor (under the same conditions) reduces to the problem of subpixel registration. For successive frames of a video camera in which there is subpixel dislocation of objects in the visual field, due to relative motion (even due to camera jitter), this provides an opportunity to increase resolution. FastVDO has developed advanced superresolution technology, as well as ATR technology, which are ready to use.
Keywords: Superresolution, Atr, Sensors, Signal Processing
