Estimation of attitude parameters from variations in image overlap

In creating geopositioned imagery from a satellite-based sensor, an image formation chain converts the raw sensor information into a form that is corrected for geometric distortions. The accuracy of these geometric corrections depends on the available ephemeris, attitude, and possibly ground control information. Unplanned sensor motion can significantly degrade the relationship between the collected attitude information and the collected image data, resulting in reduced geopositioning accuracy across the image. For multi-array push-broom imaging sensors (e.g., IRS-1C), ground areas may exist that are observed by more than one imaging array within a single image collection due to positioning (overlap) of the detector arrays on the focal plane.; This research analyzes the effects of sinusoidal motion perturbations on where ground points in the overlap region cross each detector array (sample offset) and how much time passes between the crossings (line offset). Results in this area include the identification of closed-form relationships between the spacing of the detector arrays, the frequency of the motion, the amplitude of the motion, and the alignment of the image data for sinusoidal roll motion. More complex relationships are identified for sinusoidal yaw and pitch motion. Specific circumstances are identified where unplanned motion is not observable based on these offset measurements. Also, the results showed that sample offset measurements were much less sensitive to yaw motion than to roll motion (more than an order of magnitude).; Using the relationships identified in the above analysis, algorithms are implemented to measure the offset information and to use it to identify the occurrence of unplanned motion. A frequency-based algorithm that uses the offset information is developed to correct anti-aliased attitude data to reflect unplanned motions that are related to earlier observations of IRS-1C data. The performance of this ground-based detection and correction of attitude information is evaluated against a series of simulated data sets and shown to provide attitude information that duplicates the measured offsets within ±1 pixel.