Random mapping network for tactical target reacquisition after loss of track

摘要

Abstract: Prior methods for tactical target reacquisition after aloss of track have used tree classifiers and templatematchers. Examples of prior techniques includeclassifiers that are trained with a priori data whichmakes them somewhat intolerant to temporal and dynamicchanges in the target pattern. Prior methods forreacquisition generally rely on proximity andarea-based schemes. The disadvantage of these methodsinclude their dependence on accurate and consistentsegmentation in cluttered scenarios and their need forprecise target position prediction. The random mappingnetwork algorithm (RMNRA) offers a solution using apattern memory and sparse feature matching technique.RMNRA assists the imaging tracker and improves trackingtenacity by reacquiring a tracked target after a lossof track has occurred. The reacquisition algorithm usesan associative memory to perform target patternmatching. The pattern matching technique is unique inthat it is tolerant to some of the ambiguities thatoccur with classical template pattern matchers.Weighted pattern feature vectors are stored in a memorymatrix to facilitate the matching of sensed andreference patterns dynamically over time. In addition,a sophisticated algorithm was designed to update thememory matrix over time to forget prior patterns as thetarget signature becomes stale over time and space. Thealgorithm has been implemented in real-time hardwareand flight tested with an infrared sensor. Thealgorithm is discussed and results using real IRimagery are shown.!