The main difficulty in object tracking through rapid sensor orientation change is caused by drastic changes in perceived target orientation and background. We propose to use the Lie transformation group to model and parameterize each of the major changes in sensor imagery in the course of target tracking. Simulating the biological vision system of the monkeys, the sensor imagery will be transformed and represented in perceptual system through Gabor types receptive fields. Also simulating the biological vision system, the tracking (adaptation, change countering) process will be formulated as "data driven dynamical tuning" of the receptive fields. The image tracking will be simultaneously performed in the image geometric space domain including the target orientation, and the signal space domain including signal/background polarity and contrast. Viewing the receptive field functions as the linear forms on the image space, we derive the Lie derivatives for the detection and measurement of each of component changes subjecting to the tracking.
Benefits: The proposed object tracking approach decomposes complicated multiple dimensional variations of target image characteristics to components. If successfully developed, it provides a means of high speed computing with compact processors and can be used in a computational sensor system for missiles and other military and commercial applications