Track Initialization for Multi-Static Active Sonar Systems

Abstract

The process of track initialization has two main parts: determining when a new contact of interest has appeared in the search region and computing an initial state estimate to be given to the tracking algorithm. The author has developed a new method for detecting the arrival of a new contact of interest in monostatic systems that partitions the detection level data into Hough Transform type bins. For discrete data (e.g, sonar echoes), the Hough transform is a histogram version of the discrete Radon transform where the data are partitioned into sets of non-overlapping bins at various angles that represent different possible target track trajectories. The new method tests the data in each Hough bin for the arrival of a new contact of interest with Page's test which is designed to detect changes in state by accumulating the values of an appropriate detector non-linearity. The purpose of the effort reported here is to investigate the applicability of the new method to multi-static systems. The theoretical performance predictions of the new method are quite favorable for monostatic systems; orders of magnitude reduction in the number of false tracks initialized with no reduction in target latency when compared to a popular M of N type method on a search region having realistic size and clutter density. In multi-static systems, however, the effects of high clutter density and poor data registration may increase the average latency for initializing target tracks when using the new method in a centralized architecture. When the coverage areas of two or more sensors overlap it is natural to try to employ a centralized architecture and register all of the measurements to a common frame of reference. Indeed, when the data from multiple sensors can be accurately registered superior track initialization performance can be achieved by combining all of the measurements in a single track initialization algorithm instead of using a distributed architecture that initializes tracks on each sensor separately and combines the results. When the data cannot be accurately registered it becomes more difficult to ensure that measurements from the same contact but received on different sensors will be located in the same Hough bin; the registration error may cause the measurements from different sensors to appear in different Hough bins and thereby prevent them from contributing to the same hypothesis test statistic. The problem is further compounded when the clutter density is high because the size of the Hough bins must be reduced to maintain the probability of detecting a contact of interest. Smaller Hough bins increase the probability that registration error will cause significant misalignment of the data from multiple sensors. Under those conditions it may be advantageous to initialize tracks on each sensor separately and combine the results with an appropriate track management method. This study quantifies the effects of clutter density and registration error on the proposed track initialization method to determine the conditions that require a distributed track initialization scheme. Specifically, the theoretical average target latency, at a constant average time between false alarms, for the centralized and distributed methods will be compared as a function of registration error and clutter density. In this way the combinations of registration error and clutter density that favor the centralized or the distributed architectures will be computed allowing multi-static systems designers to select the architecture appropriate for each specific application.