The Dynamical Behavior of Detected vs. Undetected Targets

Ronald Mahler, Eagan MN, USA, EML: [email protected] · 2025

Read The Dynamical Behavior of Detected vs. Undetected Targets by Ronald Mahler, Eagan MN, USA, EML: [email protected] (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2501-11189.txt.

Introduction

Ronald Mahler revisits his own past work on tracking targets that may or may not have been seen. Sensors catch some targets and miss others, and a sound filter must hold both kinds at once. His 2025 study stress-tests a known filter family that claims to run the two kinds side by side.

Mahler shows the claim cannot be made strict with two standard filter builds. A new single-step build saves part of the claim. He then derives simpler filters for both his old split and the new one. Key formulas from the old work check out by plain algebra, with no heavy stats. The upshot is a doubt on split tracking as sold, and a cleaner base to rebuild on.

Catalog record

CREATORRonald Mahler, Eagan MN, USA, EML: [email protected]
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE121 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2501-11189

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

The Dynamical Behavior of Detected vs. Undetected Targets Ronald Mahler, Eagan MN, USA, EML: [email protected] Abstract This paper is a sequel of the 2019 paper [ 5 ] . It demonstrates the following: a) the Poisson multi-Bernoulli mixure (PMBM) approach to detected vs. undetected (U/D) targets cannot be rigorously formulated using either the two-step or single-step multitarget recursive Bayes filter (MRBF); b) it can, however, be partially salvaged using a novel single-step MRBF; c) probability hypothesis density (PHD) filters can be derived for both the original “S-U/D” approach in [ 5 ] and the novel “D-U/D” approach; d) important U/D formulas in [ 5 ] can be verified using purely algebraic methods rather than the intricate statistical analysis employed in that paper; and e) the claim, that PMBM filters can propagate detected and undetected targets separately in parallel, is doubtful. 1 Introduction This paper is a sequel of the 2019 paper [ 5 ] . The primary content of that paper had two main parts: 1. In Section 4, a critique of the first three versions of the Poisson multi-Bernoulli mixture (PMBM) filter. 2. In Section 5, a statistical theory of a peripheral multitarget tracking concept that is central to PMBM filters: “undetected” vs. “detected” targets (hereafter referred to as “U-targets” vs. “D-targets”). A summary of Section 4, along with a short critique of a fourth (“trajectory”) PMBM filter incarnation, subsequently appeared in [ 7 ] . In this paper it will be demonstrated that: a) the PMBM approach to U/D-targets is seriously theoretically erroneous, but b) can be partially salvaged by adopting a novel Bayesian theoretical reformulation of it. This Introduction is organized as follows: PMBM filters (Section 1.1 ), the static model for U/D targets (Section 1.2 ),the dynamic model for U/D targets (Section 1.3 ), the five purposes of this paper (Section 1.4 ), and the organization of the paper (Section 1.5 ). 1.1 PMBM Filters Like most multitarget algorithms, PMBM filters presume detection-type sensors and the “small target” model. That is, a) at any given time t_{k} a detection process is applied to a sensor signature, resulting in a finite set Z_{k}\subseteq\mathbb{Z} of point measurements (“detections”) that are generated by targets or by clutter/false alarms; and b) single point targets can generate at most a single point detection, and target detections can be generated by only a single point target [ 11 , p. 433] . PMBM filters are also based on the random finite set (RFS) two-step multitarget recursive Bayes filter (MRBF) [ 11 , p. 484] , which recursively alternates between time-update and measurement-update steps \displaystyle f_{k|k-1}(X|Z_{1:k-1}) \displaystyle= \displaystyle\int f_{k|k-1}(X|X_{k-1})\,f_{k-1|k-1}(X_{k-1}|Z_{1:k-1})\delta X_{k-1} (1) \displaystyle f_{k|k}(X|Z_{1:k}) \displaystyle= \displaystyle\frac{f_{k}(Z_{k}|X)\,f_{k|k-1}(X|Z_{1:k-1})}{\int f_{k}(Z_{k}|X)\,f_{k|k-1}(X|Z_{1:k-1})\delta X} (2) with initial distribution f_{0|0}(X) ; and where f_{k|k-1}(X|X_{k-1}) is the multitarget Markov density and f_{k}(Z|X)=L_{Z}(X) is the multitarget measurement density; and where \int\cdot\delta X_{k-1} and \int\cdot\delta X are RFS set integrals (Eq. ( 88 )). 1.2 The Static Model for U/D Targets The purpose of Section 5 of [ 5 ] was to devise a theoretically rigorous basis for U/D targets based on their instantaneous statistical behavior—i.e., their behavior at the instant the detection process is applied. • This will be called the static U/D (S-U/D) model of U/D targets. Section 5 of [ 5 ] derived explicit formulas for the multitarget probability distributions f_{k|k}^{d}(X|Z_{1:k}) for D-targets [ 5 , Eq. (75)] and \ f_{k|k}^{u}(X|Z_{1:k}) for U-targets [ 5 , Eq. (90)] and their corresponding probability generating functionals (p.g.fl.’s) G_{k|k}^{d}(X|Z_{1:k}) , [ 5 , Eq. (5)] and G_{k|k}^{u}(X|Z_{1:k}) [ 5 , Eq. (6)] , instantaneously at observation times t_{k} . It did not consider the following question: Can individual targets be identified as U-targets vs. D-targets? 1.3 The Dynamic Model for U/D Targets PMBM filters employ a historical-behavioral , rather than an instantaneous-statistical, interpretation of U/D-targets, one which presumes that this question is answerable. Specifically, U/D targets are defined as ( [ 5 , p. 10] , lines 10-13): 1. At time t_{k} a target is a U-target if it has not been detected at any of the observation times t_{1},...,t_{k} . 2. If a U-target is detected at time t_{k+1} then it is a D-target at this and all subsequent times—i.e., D-targets cannot become U-targets. Expressed more intuitively: A target is presumed conjectural until proven “real” by the act of generating some measurement z\in\mathbb{Z} in the sensor measurement space \mathbb{Z} . However, in Item 3 of [ 5 , p. 13] and Section 9 of [ 7 , p. 3] it was noted that: 1. the PMBM filter’s U/D approach appears to violate the two-step structure of the MRBF; and 2. this incompatibility is due to the seriously erroneous physical assumption that all measurements arise from newly-detected—and therefore newly-appearing—targets. That is, they occur only when U-targets transition to D-targets. • This will be referred to as the U-birth model for measurement generation. 1.4 The Five Purposes of this Paper 1.4.1 First Purpose In Sections 2 and 4 it will be shown that, respectively, the PMBM U/D approach: 1. cannot be rigorously formulated using either the two-step MRBF or the conventional single-step “joint” version of that filter (i.e., in which the two update steps are merged into a single step, see Eqs. ( 23 , 85 )); but 2. can be rigorously reformulated using an unconventional single-step MRBF—i.e., one that employs a novel Bayes recursive transition function (see Eqs. ( 33 - 37 , 51 - 53 , 196 )). • This will be called the dynamic U/D (D-U/D) model of U/D targets. The D-U/D model consists of two parts: a) the definition of the U/D single-target state space (Eq. ( 10 )); and b) the definition of the novel Bayesian single-target recursive joint transition function (Eqs. ( 33 - 37 )). The multitarget case will be addressed in Section 6 . The D-U/D reformulation does not obviate the erroneousness of the U-birth model. For, as pointed out in [ 11 , pp. 53-55] , if measurement models are seriously inaccurate then, as signal-to-noise (SNR) decreases, a multitarget filter must increasingly “waste” data trying (and eventually failing) to overcome the mismatch between model and reality. Most obviously, measurements can also arise from clutter/false alarms. Thus when clutter is dense (SNR is small) the U-birth model will—at every observation time—result in the mandatory creation of a large number of non-existent targets. 1.4.2 Second Purpose Sections 4 - 6 and Section 7 will, respectively, a) provide a theoretically rigorous basis for the novel D-U/D model and b) compare it to the S-U/D model. The main results are as follows, assuming that the two approaches have been “aligned,” in the sense of Section 7.1 , to permit an apples-to-apples comparison: 1. D-Targets : The respective p.g.fl.’s for the S-U/D vs. D-U/D models are (see Eqs. ( 176 , 184 )): \displaystyle G_{k|k}^{d}[h] \displaystyle\propto \displaystyle\int f_{k}^{\ast}(Z_{k}|X)\,(hp_{D})^{X}\,\frac{\delta G_{k|k-1}}{\delta X}[p_{D}^{c}]\delta X (3) \displaystyle\breve{G}_{k|k}^{d}[\breve{h}] \displaystyle\propto \displaystyle\int\breve{f}_{k}^{\ast}(Z_{k}|\breve{X})\,(\breve{h}|_{1}p_{D})^{\breve{X}}\,\frac{\delta\breve{G}_{k-1|k-1}}{\delta\breve{X}}[\breve{h}^{\mathbb{D}}p_{D}^{c}]\delta\breve{X}. (4) 2. U-Targets : The respective p.g.fl.’s for the S-U/D vs. D-U/D models are (see Eqs. ( 179 , 187 )): \displaystyle G_{k|k}^{u}[h] \displaystyle\propto \displaystyle\int f_{k}^{\ast}(Z_{k}|X)\,p_{D}^{X}\,\frac{\delta G_{k|k-1}}{\delta X}[hp_{D}^{c}]\delta X (5) \displaystyle\breve{G}_{k|k}^{u}[\breve{h}] \displaystyle\propto \displaystyle\int\breve{f}_{k}^{\ast}(Z_{k}|\breve{X})\,(\breve{h}p_{D})^{\breve{X}}\,\frac{\delta\breve{G}_{k-1|k-1}}{\delta\breve{X}}[\breve{h}^{\mathbb{U}}p_{D}^{c}]\delta\breve{X}. (6) The meaning of Eqs. ( 3 - 6 ) is that the D-U/D approach appears to provide a more complete and nuanced

Read the complete text