Design and Analysis of Modern Tracking Systems / Edition 1

by Samuel S. Blackman, Robert Popoli

Overview

Here's a thorough overview of the state-of-the-art in design and implementation of advanced tracking for single and multiple sensor systems. This practical resource provides modern system designers and analysts with in-depth evaluations of sensor management, kinematic and attribute data processing, data association, situation assessment, and modern tracking and data fusion methods as applied in both military and non-military arenas.

Product Details

ISBN-13:	9781580530064
Publisher:	Artech House, Incorporated
Publication date:	08/01/1999
Series:	Artech House Radar Library Series
Pages:	1232
Product dimensions:	6.30(w) x 9.30(h) x 2.20(d)

About the Author

Samuel S. Blackman has over 35 years experience working in tracking system design. Robert F. Popoli is a teacher and has fifteen years industry experience.

Read an Excerpt

1: The Basics of Target Tracking

. . . 1.4.3 Sequential Versus Deferred Decision Logic

An ideal, batch processing, approach that would process all observations (from all time) together is typically not feasible computationally. Thus, a more standard approach has been to perform processing in a recursive (or sequential) manner as data are received. For example, a set of observations may be collected from a time frame (or scan) of a scanning radar and observation-to-track data association performed on these observations during the period when the sensor is collecting the next frame of data. Using this approach, data association decisions, once made, are irrevocable.

A deferred decision approach to data association, as exemplified by the branching algorithm of [291 and the more structured MHT method of [30], allows the final decision on difficult data association situations to be postponed until more information, such as the next frame of data, is received. In effect, this approach allows a modified version of batch processing. Alternative hypotheses are formed and evaluated when later data are received. This approach clearly has the potential for ultimately achieving a much higher correct decision probability than the sequential decision method. However, to maintain computational feasibility, an Intricate logic to delete (prune) unlikely hypotheses and to combine similar hypotheses is required.

Chapters 10 and 14 of [13] present Implementation details on Reid's algorithm [30], which was the first fully developed MHT method proposed. Chapter 6 of this book discusses MHT methods In more detail and Chapter 7 introduces the multidimensional assignment approach,which is, in effect, a form of MHT. Chapter 16 presents a detailed discussion of the implementation of a track-oriented MHT method. The book chapter by Kurien, contained in [17], also presents a detailed discussion of the implementation of a track-oriented version of MHT.

TO summarize, deferred decision logic, as exemplified by MHT, represents the most effective way to reduce the region of unstable tracking shown in Figure 1.4 for multiple closely spaced targets. Also, comparative results given in Chapter 6 of this book show that an MHT tracker can operate in clutter densities that are at least 10 times greater than the densities at which GNN can no longer successfully operate.

The all-neighbors PDA and JPDA approaches can also be implemented using a deferred decision approach. For example, a given track can still be updated with a probabilistically weighted set of observations that were received on a previous scan. However, the computation of weightings can use data received on subsequent scans. This approach is also discussed further in Chapter 7.

1.4.4 Incorporation of Group Information

Many targets of interest, particularly for military applications, tend to travel in groups (or formations). Intuitively, this information should be directly applicable for data association and track filtering. For example, even if individual tracks are maintained on aircraft In a formation, the fact that the aircraft are flying together should provide additional smoothing on the velocity estimates. Similarly, if a newly resolved target first appears close to a group, the new track formed on that target can be initiated with the group velocity.

Despite the intuitive appeal of group tracking, little development has been reported in the tracking literature since the classic work of Taenzer [311 and the overview discussion given in Chapter I I of [ 131. It appears that the implementation problems resulting from the splitting and merging of groups have discouraged the further development of group tracking methods. However, it seems clear that some form of group tracking is the best approach for the tracking environment of many closely spaced targets shown in the lower region of Figure 1.4. Thus, Chapter 6 presents a brief update to the discussion in Chapter 11 of [ 13] and Chapter 16 outlines an approach that combines MHT and group track methodologies.

1.4.5 Use of Thresholding

Until now, the discussion has implicitly assumed that the measurement process, shown in Figure 1.2, will include detection thresholding so that a set of observations can be produced. This observation set would include only the output from detection bins, such as pixels in an IR sensor or range/range rate bins in a radar, in which the measured signal intensity, amplitude, or SNR exceeded the detection threshold.

Although there may be some feedback from the tracker to the signal processing system, the use of detection thresholding clearly separates the functions of detection and tracking. Also, thresholding leads to an irretrievable loss of information whereby, for example, a signal just below the threshold level is completely ignored. Thus, as an alternative to thresholding, an approach has been proposed that considers the detection and track confirmation processes to occur, in effect, simultaneously [321. As stated in [321, for this approach "the detection of a target and the estimation of its state are an inseparable part of the same decision process." This approach can be contrasted with the traditional sequential approach whereby a detection device is used to generate observations and then track initiation and confirmation logic operates on these observations to form tracks. This alternative approach is frequently denoted track before detect (TBD), although, in effect, the tracking and detection processes occur simultaneously.

The TBD approach relies on a mapping from the measurement space to the target state space. Although the target moves in measurement space, its position in a suitable chosen state space may remain relatively constant. Thus, the signal power, or intensity, in the target state space can be summed over time to determine target presence. For example, the approach of [321 maps (and integrates) measured intensity into "bins" representing initial target position and velocity, which are assumed constant. Alternatively, for an IR system, four-dimensional bins consisting of two angles and the corresponding angle rates can be defined. Then, the likelihood that a target exists in a given bin can be computed using the current measured signal intensity in that bin and the transition probabilities from other adjoining bins that previously had, on the past scan, an appreciable likelihood of target presence.

The practical problem associated with the TBD approach is the necessity to search many regions, or "bins," of target state space. The example discussed in [32] considers only two dimensions and assumes essentially constant target velocity. Still, a very extensive search is indicated. Reference [331 considers a similar approach using neural nets in which each neuron represents a bin in target state space. However, again the problem is greatly simplified by the assumption of targets moving in a two-dimensional space. Thus, the computational requirements for TBD appear to be formidable. As an example of TBD computational requirements, simple calculations indicate the potential for a requirement of up to about 1013 bins to cover a three-dimensional position and velocity space and to provide accuracy comparable to that of current conventional systems. On the other hand, the TBD approach, whereby bins in target state space are examined for target presence, has the computational advantage that parallel processing techniques are directly applicable. The same basic calculations are performed on all bins regardless of the number of targets present or the level of the clutter density. This feature is in distinct contrast to the MHT approach in which dim targets may be detected by using lower thresholds but where the potential growth in the number of tracks and hypotheses that must be maintained may make the implementation unfeasible, even for modern computational systems.

Numerous studies, including Arnold et al. [34, 35] and the chapter by Barniv in [171, have shown that the dynamic programming algorithm (DPA) provides a particularly efficient means to integrate signal strength/likelihood over time, and through bins in target state space, without thresholding. Shaw and Arnold [35] also show that this TBD approach can be used to form track segments that are connected into continuous tracks using MHT. Thus, the combination of TBD and MHT might be the best practical solution to the problem of tracking dim targets....

Table of Contents

The Basics of Target Tracking. Sensor and Source Characteristics. Kinematic State Estimation: Filtering and Prediction. Modelling and Tracking Dynamic Targets. Passive Sensor Tracking. Basic Methods for Data Association. Advanced Methods for MTT Data Association. Attribute Data Fusion. Multiple Sensor Tracking — Issues and Methods. Multiple Sensor Tracking — System Implementation and Applications. Reasoning Schemes for Situation Assessment and Sensor Management. Situation Assessment. Tracking System Performance Prediction, and Evaluation. Multi Target Tracking with an Agile Beam Radar. Sensor Management. Multiple Hypothesis Tracking System Design and Application. Detection and Tracking of Dim Targets in Clutter.

Editorial Reviews

Discusses the basics of target tracking; sensor and source characteristics; filtering and prediction in kinematic state estimation; modeling and tracking dynamic targets; passive sensor tracking methods; basic methods for data association; advanced methods for MTT data association; attribute data fusion; issues, methods, system implementation, and applications of multiple sensor tracking; reasoning schemes for situation assessment and sensor management; situation assessment; tracking system performance prediction and evaluation; multiple target tracking with an agile beam radar; sensor management; multiple hypothesis tracking system design and application; and detection and tracking of dim targets in clutter. Annotation c. Book News, Inc., Portland, OR (booknews.com)

Booknews