Downloads & Free Reading Options - Results

Standard tracking algorithms involve processing measurements associated with a given target and forming an estimate of the target's state. However, many practical situations also include an uncertainty regarding the origin of the data. In the most general case one is faced with the problem of tracking targets in a multinsensor multitarget environment, possibly including highly dissimilar data types ranging from acoustic measurements to visual sightings. The term data association, as applied here, refers to the partitioning of a set of measurements according to their sources. Each such partition is termed a hypothesis, and the object is to find the best hypothesis. In this report the authors describe an algorithm which is intended to provide a general framework for data association. It is cast in a Bayesian context; that is, the relative merits of the hypotheses are evaluated in terms of their aposteriori probabilities. However, the presentation includes the development of a general data and scoring structure which should find application in most schemes which evaluate hypotheses recursively. In addition, a detailed model is provided for the case of bearings-only measurements in a convergence zone environment. The methodology used in developing this restricted case is considered a prototype for future models. (Author)