Track before Detect Algorithms
Abstract
Track before detect(TBD)is an algorithm to process the received signal directly. The algorithm can get exact detection and track at the same time, Classical processing is to implement tracking algorithm after threshold processing. Comparing to the classical tracking algorithm, TBD fully uses the information hiding in measurement. So it’s a useful algorithm to detect and track target in low signal noise ratio environment. This article mainly studies how to carry out TBD by dynamic programming algorithm and particle filter. Then the principle and the application of over the horizon radar(OTHR)are introduced. After giving the model of TBD of OTHR, the two kinds of algorithms are used to carry out the TBD of OTHR. Mainly content of the article is: introduce dynamic programming and particle filter based TBD, including their theory and Performance Comparison, the framework of the algorithms, the using bounds, their advantages and disadvantages.
Keywords: maneuvering weak target, track before detect track dynamic programming, algorithm, particle filter
1
Seamless detection and tracking schemes are able to integrate unthresholded (or below target detection threshold) multiple sensor responses over time to detect and track targets in low signal-to-noise ratio (SNR) and high-clutter scenarios. Moreover, iterative tracking algorithms must be initiated appropriately. Under simple conditions, this is not a difficult task. For low observable objects, that is, for targets embedded in a high false return background or in case of incomplete measurements, more than a single frame of observations are usually necessary for detecting the existence of all objects of interest moving in the sensors' fields of view. The tracking iteration is then initiated by "extracted" target tracks, that is, by tentative tracks whose existence is "detected" by a detection process on a higher level of abstraction making use of sensor detections accumulated over time.
These schemes, also called "track-before-detect (TBD)" algorithms, are especially suitable for tracking low-observable targets that would only very rarely cross a standard detection threshold as applied at the sensor level.
Containing 11 papers, this special issue presents latest research in TBD. The topics cover a wide range of research areas including TBD theories and algorithms, including comparison of their performances, and TBD applications in various surveillance scenarios: sequences of infrared images, radar data, infrared radar images, tracking and classification with image and video-based sensors, airborne QuickSAR, and blind mobile terminal position tracking. Theory for TBD
In the first paper in this issue, T. A. Wettergren and M. J. Walsh derived analytical expressions for the expected value and variance of the area of uncertainty achieved by employing a track-before-detect search strategy for localizing a target moving across a distributed sensor network. The analytical expressions were verified by comparison with computational experiments showing exemplary scenarios of uniform, barrier, and arbitrary field designs. In the second paper, M. McDonald and B. Balaji processed real radar data using a finite difference (FD) implementation of continuous-discrete filtering with a four-dimensional constant velocity model. Measurement data is modeled assuming a Rayleigh distributed sea clutter with embedded Swerling 0, 1, or 3 target signal models. The results are examined to obtain a qualitative understanding of the effects of using the different target models. The Swerling 0 model is observed to exhibit a heightened sensitivity to changes in measured signal strength and provides enhanced detection of the maritime target examined at the cost of more peaked or multimodal posterior density in comparison with Swerling 1 and 3 targets.
TBD Algorithms
C. R. Berger et al. introduced an analytical approach to initialize a tracking filter from a minimum number of observations. This directly pertains to multihypothesis tracking (MHT), where in the presence of clutter and/or multiple targets (i) a low-complexity algorithm is desirable and (ii) using a small set of measurements avoids the combinatorial explosion. Two different 2
implementations are compared, differing in the approximation of the posterior: linearizing the measurement equation as in the extended Kalman filter (EKF) or employing the unscented transform (UT). The approach has been studied in practical examples: 3D track initialization using bearings-only measurements or using slant-range and azimuth only. For these examples, the authors provide detailed discussion and numerical analysis of the examples using Cramer-Rao lower bounds and Monte Carlo simulation for performance comparison and proof of consistency.
O. Nichtern and S. R. Rotman addressed the problem of tracking a dim moving point target in a sequence of IR images. A dynamic programming algorithm (DPA) is used to process the frames of an image sequence containing a target in low SNR conditions. The paper deals with the practical issues of setting the parameters of their algorithm to maximize tracking capability depending on the different levels of noise and different target velocities and mobilities. Therefore, the algorithm is applied to real datasets.
M. Wieneke and W. Koch proposed an integration of a sequential likelihood-ratio (LR) test into the probabilistic multiple hypothesis tracking (PMHT) framework.
An LR formula for track extraction and deletion using the PMHT update formulae is presented. The resulting update formula for the sequential LR test affords the development of track-before-detect algorithms for PMHT. The approach is illustrated by a simulation example.
Performance Comparison
W. Blanding et al. extended the maximum likelihood-probabilistic data association target tracking (ML-PDA) algorithm from a single-target tracker to a multitarget tracker and compare its performance to that of the probabilistic multihypothesis tracker (PMHT). Before developing the extension to the multitarget tracker, they described recent advances in ML-PDA which make it suitable for real-time tracking. Advances are made in computational efficiency and reliability in track validation.
S. J. Davey et al. compared, on the basis of detection performance and computation resource requirements, four different implementations of the TBD paradigm: Bayesian estimation over a discrete grid, dynamic programming, particle filtering methods, and the histogram probabilistic multihypothesis tracker. Similarities and differences, that they found, are discussed and explained.
Applications
M. G. S. Bruno et al. implemented the track-before-detect methodology by sequential Monte Carlo methods to infrared radar images. Their method enables integrated, multiframe target detection and tracking incorporating the statistical models for target aspect, target motion, and background clutter. Two implementations of the proposed algorithm are discussed using, respectively, a resample-move (RS) particle filter and an auxiliary particle filter (APF). Their simulation results suggest that the APF configuration outperforms slightly the 3
RS filter.
M. Asadi and C. S. Regazzoni developed a novel method to track nonrigid objects in the presence of occlusion. Objects and their dynamic shape are described by a set of corners. Tracking needs to take place in a multimodal voting space where multimodality occurs because of occlusion events and clutter. The method implements a model-based learning strategy. The approach has been tested with several video sequences showing pedestrians occluded either by a car or another pedestrian. Experimental results show a successful tracking along with a more precise estimation of shape and motion during occlusion events.
P. K. Sanyal et al. addressed the important problem of detecting and georegistering surface moving targets in multichannel synthetic aperture radar (SAR) by an interferometric scheme which allows to detect moving targets well within ground clutter. By multiple threshold comparisons and grouping of pixels within the intensity and the phase images, they showed how to reliably detect and accurately georegister moving targets within short duration SAR. Furthermore, they described a novel channel-to-channel clutter cancellation technique that enhances the performance for the detection of moving targets. The new techniques are applied to real multichannel radar data and result in good performance.
Finally, V. Algeier et al. presented a track-before-detect method for initialization of blind mobile terminal tracking in urban scenarios. The method explicitly takes advantage of multipath propagation. The urban propagation channel is modeled by using context information about the location of the main scattering objects such as buildings which enters into a real-time ray tracing technique. By this, the underlying measurement likelihood function can algorithmically be calculated for a randomly distributed set of potential transmitter positions, which explains the measurement with respect to the main propagation channels. The likelihood function is the key component of a TBD scheme providing initial state estimates for mobile transmitter tracking using particle filtering techniques. The simulation results illustrate the potential of the method.
4
References
[1] Felix Govaers, Yang Rong, Lai Hoe Chee et Journal on Advances in Signal Processing, 2011, 2011(1). Journal on Advances in Signal Processing, 2008, 2008.
[3] Yvo Boers, Frank Ehlers, Wolfgang Koch et Journal on Advances in Signal Processing, 2008, 2008(1).
[4] A. Hemmingsson, B. Jung, C.
[5] Otto of the American Musicological Society, 1937(2).
[7] Harry G. of Political Economy, 1972, 80(3).
[9] Laukkanen A.-M., Kankare Phoniatrica et Logopaedica, 2006, 58(4). Journal on Advances in Signal Processing, 2008, 2008.
硕研——16班
精密仪器及机械
S140xxxxxxxx
王英
5