Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Revisiting Functional Derivatives in Multi-object Tracking

Jan Krejčí, Ondřej Straka, Petr Girg, Jiří Benedikt

TL;DR

This work tackles the mathematical rigor of functional derivatives used in multi-object tracking (MOT) based on random finite sets (RFS) and probability generating functionals (PGFLs). It shows that existing derivative definitions often fail to guarantee a chain rule or require restrictive spaces, and then introduces a measure-based redefinition—PGFM—where the input is a complex Radon measure and the Dirac delta is treated as a measure. The framework proves that first- and higher-order derivatives of the PGFM are Fréchet, yielding density and factorial-moment expressions as natural special cases, and establishing the correct differentiation rules for MOT applications. Open problems remain around practical implementability, order-dependence of derivatives, and extensions to more general spaces, but the approach provides a rigorous foundation for deriving MOT algorithms with robust mathematical guarantees.

Abstract

Probability generating functionals (PGFLs) are efficient and powerful tools for tracking independent objects in clutter. It was shown that PGFLs could be used for the elegant derivation of practical multi-object tracking algorithms, e.g., the probability hypothesis density (PHD) filter. However, derivations using PGFLs use the so-called functional derivatives whose definitions usually appear too complicated or heuristic, involving Dirac delta ``functions''. This paper begins by comparing different definitions of functional derivatives and exploring their relationships and implications for practical applications. It then proposes a rigorous definition of the functional derivative, utilizing straightforward yet precise mathematics for clarity. Key properties of the functional derivative are revealed and discussed.

Revisiting Functional Derivatives in Multi-object Tracking

TL;DR

This work tackles the mathematical rigor of functional derivatives used in multi-object tracking (MOT) based on random finite sets (RFS) and probability generating functionals (PGFLs). It shows that existing derivative definitions often fail to guarantee a chain rule or require restrictive spaces, and then introduces a measure-based redefinition—PGFM—where the input is a complex Radon measure and the Dirac delta is treated as a measure. The framework proves that first- and higher-order derivatives of the PGFM are Fréchet, yielding density and factorial-moment expressions as natural special cases, and establishing the correct differentiation rules for MOT applications. Open problems remain around practical implementability, order-dependence of derivatives, and extensions to more general spaces, but the approach provides a rigorous foundation for deriving MOT algorithms with robust mathematical guarantees.

Abstract

Probability generating functionals (PGFLs) are efficient and powerful tools for tracking independent objects in clutter. It was shown that PGFLs could be used for the elegant derivation of practical multi-object tracking algorithms, e.g., the probability hypothesis density (PHD) filter. However, derivations using PGFLs use the so-called functional derivatives whose definitions usually appear too complicated or heuristic, involving Dirac delta ``functions''. This paper begins by comparing different definitions of functional derivatives and exploring their relationships and implications for practical applications. It then proposes a rigorous definition of the functional derivative, utilizing straightforward yet precise mathematics for clarity. Key properties of the functional derivative are revealed and discussed.

Paper Structure

This paper contains 44 sections, 5 theorems, 82 equations, 1 table.

Table of Contents

  1. Introduction
  2. Background on RFS Modeling
  3. The Base Space
  4. Brief Introduction to RFSs
  5. Belief Mass Function
  6. Density Function
  7. Set Integral
  8. Probability Generating Functional
  9. Functional Derivatives in MOT
  10. PGFL $\rightarrow$ density function
  11. PGFL $\rightarrow$ factorial moment
  12. Problem Statement
  13. State of the Art in Functional Derivatives: Definitions and Analysis
  14. Definition Bypassing Limit Procedure
  15. Definition Almost Everywhere
  16. ...and 29 more sections

Key Result

Proposition 1

The Dirac measure is a Radon measure assuming $\mathcal{X}$ is locally compact Hausdorff (lcH) space, i.e., $\delta_{\mathbf{x}} \in M(\mathcal{X})$. The proof is given in Appendix appendix:Dirac_deltas-proof. $\square$

Theorems & Definitions (5)

  • Proposition 1
  • Proposition 2
  • Corollary 3
  • Corollary 4
  • Theorem 5