Detection and Tracking as a Seamless Process
Abstract
Detection and tracking are normally considered separate processes. First the signal processing system associated with a sensor examines the signal to determine whether to call detection. Once detection is called, it is converted to an estimate of one or more of the components of the target's kinematic state, e.g., bearing, position, or velocity. This estimate (contact) is sent to a tracking system that determines whether the contact should be associated with an existing track or used to generate a new one. This process works well in high signal-to-noise ratio (SNR) situations but sacrifices performance in low ones. The tracking community is making progress toward seamless detection and tracking and recovering some of this lost performance. In this talk, we present a method called likelihood ratio detection and tracking (LRDT) that is a step toward providing an integrated approach to detection and tracking. We provide examples of the application of LRDT to sonar and radar detection and tracking LRDT is a recursive Bayesian version of track-before-detect. In LRDT one specifies a surveillance region that has a prior probability less than one of containing a target. There is a probabilistic motion model that specifies target motion within the region as well as the possibility of transiting into and out of the region. As sensor information is received, it is converted into a measurement likelihood ratio function and combined with the prior likelihood ratio surface to produce a posterior surface. Peaks in this surface are used to determine whether a target is present and to provide an estimate of its state (track). The process is recursive with the posterior surface from one time period being updated for target motion to become the prior for the next measurement likelihood ratio function.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 17, 2004
- Accession Number
- ADA432583
Entities
People
- Lawrence Stone