Time-Sequential Optical Data Processing

摘要

Our work has emphasized analog optical processing for three levels of multitarget tracking: target detection, track initiation and target tracking. New algorithms and new optical processing architectures have resulted and initial quantitative results have been obtained for each phase of the work. A new acousto-optic implementation of our sub-pixel target detection algorithm was devised and tested. This involved a new algorithm and architecture that achieves image registration, interpolation and subtraction. This technique is thus suitable for general image registration applications also. A 2-D Hough transform optical processors is used for track initiation. It allows accumulation of target measurements until sufficient confidence has been established that a target track exists. It provides position and velocity estimates for each target and predictions of the next vector location of each target. A new threshold Hough transform provides improved performance and reduced false alarms in noise. Very impressive results from many multiple targets were obtained. This work is also suitable for general object feature extraction and product inspection applications. We devised a completely new realization of the Joint Probabilistic Data Association Algorithm for multitarget tracking and a most attractive new acousto-optic architecture to achieve this. This also has more general use in hypothesis testing. Finally, a new analogy neural net algorithm and architecture was formulated, an optical processor to implement it was described and initial simulations of it were obtained. (fr)