Pith. sign in

REVIEW 2 major objections 4 minor 38 references

Revisiting Split Covariance Intersection: Correlated Components and Optimality

T0 review · 2 major / 4 minor · reviewed 2026-08-10 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The paper proves that Extended Split Covariance Intersection gives the smallest possible conservative covariance bound when fusing two estimators with known correlated error components.

desk verdict Solid new fusion rule with a genuinely new optimality proof under positive-definite covariances, but the claimed extension to the PSD case rests on a false equivalence and needs fixing. read the letter →

arxiv 2501.07915 v1 pith:4IS7GVC7 submitted 2025-01-14 eess.SP

Colin Cros (GIPSA-INFINITY , GIPSA-GAIA) , Pierre-Olivier Amblard (GIPSA-GAIA) , Christophe Prieur (GIPSA-INFINITY) , Jean-François Da Rocha This is my paper · ORCID
classification eess.SP MSC 93E1093E1115A45
keywords conservativefusioncovarianceintersectionsplitdistributedestimationlinearellipsoidboundscross-covarianceoptimal
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper studies linear fusion of two estimators whose error covariances are only partially known: each error splits into a component whose cross-covariance with the other estimator is unknown and a component whose covariances and cross-covariances are known. Its central claim is that a fusion rule called Extended Split Covariance Intersection (ESCI) yields the smallest possible conservative upper bound on the fused error covariance, in the sense of any increasing cost function, over the whole set of admissible covariances. This matters because distributed estimation algorithms must fuse estimates without knowing the correlations induced by shared process noise, and any bound that underestimates the error can be dangerous. If the claim is correct, safe fusion of two estimators reduces to choosing a single weight on the segment $[0,1]$, rather than optimizing over a high-dimensional set of covariance matrices, and the known optimality of Covariance Intersection is recovered as a special case.

What carries the argument

The central object is the ESCI centralized bound $B_c^{\mathrm{ESCI}}(\omega)=\operatorname{diag}(\omega^{-1}\tilde P_1^{(1)},(1-\omega)^{-1}\tilde P_2^{(1)})+\tilde P_c^{(2)}$ and its fused ellipsoid $B_F^{\mathrm{ESCI}}(\omega)=(H^\top B_c^{\mathrm{ESCI}}(\omega)^{-1}H)^{-1}$, with $\omega\in[0,1]$ a single fusion weight. The argument is carried by three pieces: the minimal volume $V(\mathcal{A}_{\mathrm{ESCI}})$ that any conservative bound must contain; the strict concavity of the function $h_x(\omega)=x^\top A_F^{\mathrm{ESCI}}(\omega)x$, which ensures the maximum over $\omega$ is attained uniquely; and a tight-circumscription theorem, adapted from the classical characterization of the intersection of two ellipsoids, showing only ESCI ellipsoids tightly circumscribe $V(\mathcal{A}_{\mathrm{ESCI}})$. In the common-noise case, a Woodbury-inversion form of the same bound reduces the computational cost from $O(N^3d^3)$ to $O(Nd^3)$.

What would settle it

Take $d=2$ with explicit matrices satisfying the paper's setup and inspect $\mathcal{A}_{\mathrm{ESCI}}(\varepsilon)$: if there is a matrix $P'_c=P_c^{(1)}+\tilde P_c^{(2)}+2\varepsilon I$ whose block $P_c^{(1)}-\varepsilon I$ is not positive semidefinite, then it has no preimage under the map $P_c\mapsto P_c+2\varepsilon I$, contradicting the general-case proof. Exhibiting such a matrix would show that Theorem 1 as stated is not established for positive-semidefinite covariances.

Watch

Extended reading notes

Core claim

The paper's main theorem states that a conservative fusion $(K,B_F)$ solves the optimal conservative fusion problem for two estimators if and only if $B_F$ equals $B_F^{\mathrm{ESCI}}(\omega^*)$ for some weight $\omega^*$ minimizing the increasing cost $J$ among the ESCI bounds. In particular, ESCI itself is always a solution. The proof characterizes the minimal volume that every conservative bound must contain: the union over admissible centralized covariances of the optimal-fusion ellipsoids. It proves that this volume can also be written as the intersection of the ESCI ellipsoids, and that the only ellipsoids tightly circumscribing it are ESCI ellipsoids. Together these facts rule out any smaller conservative bound, so the ESCI family is optimal for two estimators regardless of which increasing cost function is used.

Load-bearing premise

The proof of the positive-semidefinite case rests on the claim that every admissible covariance in the perturbed set $\mathcal{A}_{\mathrm{ESCI}}(\varepsilon)$ can be written as some admissible covariance plus $2\varepsilon I$; in dimension $d>1$ this mapping can fail to be onto because the perturbed first-component block may no longer be positive semidefinite, so the general theorem depends on Assumption 1, or on a corrected limiting argument.

Editorial extensions

If this is right

  • For two estimators, any conservative fusion that is optimal for some increasing cost function is an ESCI fusion, and the search for the best bound is a one-dimensional optimization over $\omega\in[0,1]$.
  • ESCI reproduces Covariance Intersection, Split Covariance Intersection, Partitioned Covariance Intersection, and the fully known-covariance optimal fusion as special cases, so their optimality properties are unified by Theorem 1.
  • In distributed estimation with a common process noise, ESCI gives smaller guaranteed error bounds than SCI or CI, with simulation gains around 20 percent for position in the paper's example, and with no additional communication or computation cost.
  • Because the bound is optimal for every increasing cost function, the choice between criteria such as trace or determinant does not affect the existence of an ESCI solution, though the optimizing weight can differ.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Inference: the one-parameter structure of the ESCI family suggests that for low-dimensional states the optimal weight $\omega$ for trace or determinant might be expressible in closed form, extending existing closed-form results for Covariance Intersection; the paper leaves this to future work.
  • Inference: the minimal-volume/tight-ellipsoid technique is likely transferable to other partial-knowledge fusion settings, such as element-wise known correlation blocks, where the admissible set is defined by inequalities rather than fixed diagonal blocks; the paper does not explore this.
  • Inference: the theorem's limitation to two estimators is structural, not technical: once a fused estimate from a previous iteration enters, the error splits into more than two unknown-correlation components, so extending ESCI to multi-step distributed algorithms would require a different optimality argument.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

2 major / 4 minor

Summary. The paper introduces Extended Split Covariance Intersection (ESCI), a conservative linear fusion rule for two estimators whose errors are split into a first component with unknown cross-covariances and a second component with known second-order moments, including correlations. The rule unifies CI, SCI, and PCI as special cases. The central theoretical claim is Theorem 1: for two estimators, a conservative fusion is optimal if and only if its bound coincides with an ESCI bound parametrized by a weight ω that minimizes the chosen cost. The proof strategy follows Reinhardt et al.: it characterizes the minimal volume V(AESCI) that every conservative bound must contain, proves in Theorem 3 that this volume equals the intersection of the ESCI ellipsoids, proves in Theorem 4 that ESCI ellipsoids are the only tight circumscribers of this volume, and then transfers this geometric tightness to cost optimality. Section 5 applies ESCI to a distributed tracking problem and reports tighter bounds than CI and SCI. The derivation under the positive-definiteness Assumption 1 is detailed and appears internally consistent; the main weakness is the attempted extension to positive-semidefinite covariances in Section 4.5, which contains an incorrect equivalence.

Significance. If fully established, Theorem 1 is a meaningful generalization of the optimality of Covariance Intersection and would provide a theoretical justification for using ESCI in distributed estimation with common process noise. The geometric proof under Assumption 1 is self-contained, uses no fitted parameters, and yields a reusable characterization of the minimal volume via Theorem 3. The application section demonstrates concrete improvements and notes that ESCI requires no additional communication compared with SCI. The significance is currently tempered by the fact that the positive-semidefinite extension of the main theorem relies on a false identification of the regularized admissible set, so the unqualified statement of Theorem 1 is not yet proven.

major comments (2)
  1. [Section 4.5] The proof of the general positive-semidefinite version of Theorem 1 relies on the assertion that membership of Pc in AESCI is equivalent to membership of Pc + 2εI_{2d} in AESCI(ε). Only the forward implication is true. For example, with d=1, P̃^(1)_1 = P̃^(1)_2 = 1, P̃^(2)_c = 0, and ε = 0.1, the matrix Q = [[1.2, 1.05], [1.05, 1.2]] belongs to AESCI(ε), but Q − 0.2I = [[1, 1.05], [1.05, 1]] is indefinite and hence not in AESCI. Consequently, the step asserting that if all KPcKᵀ lie below BF for AESCI then all KPε_cKᵀ lie below BF + 2εKKᵀ for AESCI(ε) does not follow, and Theorem 4 cannot be applied to BF + 2εKKᵀ. Thus Theorem 1 as stated is not established for positive-semidefinite covariances. Please either restrict Theorem 1 and Corollary 2 to Assumption 1 or replace the argument by a valid limiting procedure.
  2. [Section 4.5] The if-and-only-if statement in Theorem 1 presumes that J is strictly increasing in the Loewner order. The proof only uses that J is increasing, and the displayed inequalities give J(BESCI_F(ω*)) ≤ J(BESCI_F(ω1)) ≤ J(BF). For a merely nondecreasing J, a bound BF different from BESCI_F(ω1) can attain the same cost and would be a solution of Problem 2 without being of the ESCI form. The manuscript should define the class of admissible cost functions and state the strict monotonicity assumption used in the characterization of all solutions.
minor comments (4)
  1. [References] Reference [37] contains a typo: 'ransactions' should be 'Transactions'.
  2. [Section 3.2] The reduction to P̃^(1,2)_c = 0 via (17) assumes that P̃^(2)_c is invertible; outside Assumption 1 this is not guaranteed and the text should at least mention how singular cases are handled.
  3. [Section 4.4, Case 3.2] In the subcase where χ(λ) leaves [0,1] in every neighborhood of 0, the text spells out the argument for ω0 = 0 and says the case ω0 = 1 is symmetric; spelling out the symmetric argument would improve readability.
  4. [Section 5.3] The sentence reporting that ESCI bounds are 'about 20% lower' for position, '5% lower' for velocity, and '1% lower' for acceleration would benefit from stating whether these are averaged over nodes, over time, or both.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: ESCI's optimality is derived from the admissible-set definition and general conservative-fusion constraints, with only a disclosed antecedent self-citation.

full rationale

The derivation chain is self-contained and does not reduce to its own inputs. The paper starts from Problem 2, which optimizes over all conservative fusion gains and bounds satisfying BF >= K Pc K^T for every Pc in the admissible set AESCI. The ESCI family is then introduced as one particular parametric construction. The optimality proof does not assume the conclusion: Lemma 4 forces every conservative bound to contain the minimal volume V(AESCI); Theorem 3 characterizes V(AESCI) by minimizing over admissible covariances Pc in AESCI; Theorem 4 shows that any ellipsoid tightly circumscribing V(AESCI) must be an ESCI ellipsoid; and Theorem 1 then orders an arbitrary conservative bound against the ESCI family. None of these steps is a rewording of the definition of ESCI: the key equality V(AESCI) = intersection of ESCI ellipsoids is proved rather than asserted. The weight omega is an optimization variable chosen by the cost function J, not a parameter fitted to data, so no fitted quantity is renamed as a prediction. The only self-reference is the disclosed antecedent work [10], used to say that ESCI extends that preliminary construction, not to justify the optimality theorem. The positive-definite case is proved directly from matrix lemmas, and the paper's Section 4.5 extension to positive-semidefinite covariances may contain a technical proof gap, but that is a correctness concern, not circularity. The central claim therefore has independent mathematical content.

Assumptions & free parameters 0 free parameters · 5 assumptions · 0 invented entities

The central result is a mathematical derivation, so the ledger is small. The key modeling input is the split-error information structure A_ESCI; the proof then uses standard matrix-analysis tools and Assumption 1. The removal of Assumption 1 is where the derivation has a gap, and the strictness of the cost function is an unstated assumption.

assumptions (5)
  • domain assumption The estimation errors can be split as x̃_i = x̃_i^(1) + x̃_i^(2) where the cross-covariances of the first components are unknown and the full second-order statistics of the second components, including cross-covariances between second components, are known.
    This defines the admissible set A_ESCI in (18) and is the information model on which the fusion and its optimality are built. It is motivated by the distributed estimation example in Section 3.1, where the common process noise creates known correlations.
  • domain assumption Positive definiteness of P̃^(1)_1, P̃^(1)_2, and P̃^(2)_c (Assumption 1) is needed for Lemmas 5, 6, and Theorem 3.
    Assumption 1 is stated in Section 4.2. The paper claims it is relaxed in Section 4.5, but the relaxation proof has a gap because the asserted equivalence between A_ESCI and A_ESCI(ε) does not hold in general.
  • standard math Standard matrix-analysis results: Woodbury inversion identity, Schur complement / Horn-Johnson Lemma 7.7.6, the Implicit Function Theorem, and Kahan's ellipsoid circumscription result.
    These are invoked in the proofs of Lemmas 3, 5, 6, and Theorem 4. They are external standard tools and not part of the claimed contribution.
  • domain assumption The cost function J is increasing in the Loewner order, implicitly strictly so for the 'if and only if' direction of Theorem 1.
    Problem 2 and Theorem 1 in Section 4.1 use an 'increasing' cost function. The proof that equality of J implies equality of the bound requires strict monotonicity, which is not stated.
  • domain assumption In the common-noise application, the process noise covariance Q and the matrices M_i are exactly known, and nodes can reconstruct all needed covariances from transmitted covariance matrices.
    This is used in Section 5.3 for the communication-cost argument. If Q or M_i are uncertain, the ESCI bound is no longer guaranteed to be conservative.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Revisiting Split Covariance Intersection: Correlated Components and Optimality." pith.science (2026). https://pith.science/paper/4IS7GVC7

@misc{pith2026250107915,
  author       = {Pith},
  title        = {Pith review of: Revisiting Split Covariance Intersection: Correlated Components and Optimality},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4IS7GVC7}},
  note         = {Machine review of arXiv:2501.07915}
}
read the original abstract

Linear fusion is a cornerstone of estimation theory. Implementing optimal linear fusion requires knowledge of the covariance of the vector of errors associated with all the estimators. In distributed or cooperative systems, the cross-covariance terms cannot be computed, and to avoid underestimating the estimation error, conservative fusions must be performed. A conservative fusion provides a fused estimator with a covariance bound that is guaranteed to be larger than the true, but computationally intractable, covariance of the error. Previous research by Reinhardt \textit{et al.} proved that, if no additional assumption is made about the errors of the estimators, the minimal bound for fusing two estimators is given by a fusion called Covariance Intersection (CI). In distributed systems, the estimation errors contain independent and correlated terms induced by the measurement noises and the process noise. In this case, CI is no longer the optimal method. Split Covariance Intersection (SCI) has been developed to take advantage of the uncorrelated components. This paper extends SCI to also take advantage of the correlated components. Then, it is proved that the new fusion provides the optimal conservative fusion bounds for two estimators, generalizing the optimality of CI to a wider class of fusion schemes. The benefits of this extension are demonstrated in simulations.

Figures

Figures reproduced from arXiv: 2501.07915 by the authors.

Figure 1
Figure 1. Comparison of the CI, SCI and ESCI bounds for the fusion of two estimators whose errors [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Comparison of the minimal volumes of CI, SCI and ESCI. The minimal volumes [PITH_FULL_IMAGE:figures/full_fig_p010_2.png] view at source ↗
Figure 3
Figure 3. Visualization of equality (36) in the case of [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Example of tight ellipsoid. The ellipsoid [PITH_FULL_IMAGE:figures/full_fig_p012_4.png]
Figure 5
Figure 5. Figure 5: Network considered. The measurement performed by the nodes are given in brackets. 5 Application to distributed es￾timation To illustrate the advantages of the new ESCI fusion rule over the usual SCI rule, this section presents its appli￾cation to a distributed estimati…
Figure 7
Figure 7. Figure 7: Estimated variance bounds (matte curves) and MSEs (semi-transparent curves) of the estimators [PITH_FULL_IMAGE:figures/full_fig_p017_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

38 extracted references · 37 canonical work pages

  1. [37]

    Z. Wu, Q. Cai, and M. Fu. Covariance intersection for partially correlated random vectors. ransactions on Automatic Control , 63(3):619–629, 2017

    work page 2017

  2. [1]

    Ajgl and O

    J. Ajgl and O. Straka. Fusion of multiple estimates by covariance intersection: Why and how it is subop- timal. International Journal of Applied Mathematics and Computer Science , 28(3):521–530, 2018

    work page 2018

  3. [2]

    Ajgl and O

    J. Ajgl and O. Straka. Rectification of partitioned covariance intersection. In American Control Con- ference, pages 5786–5791. IEEE, 2019

    work page 2019

  4. [3]

    Ajgl and O

    J. Ajgl and O. Straka. Covariance intersection fusion with element-wise partial knowledge of correlation. Automatica, 139:110168, 2022

    work page 2022

  5. [4]

    Ajgl and O

    J. Ajgl and O. Straka. Linear fusion with element- wise knowledge. In 25th International Conference on Information Fusion, pages 1–8. IEEE, 2022

    work page 2022

  6. [5]

    B. D. O. Anderson and J. B. Moore. Optimal fil- tering. Prentice-Hall, Inc., Englewood Cliffs, N.J. 07632, 1979

    work page 1979

  7. [6]

    P. O. Arambel, C. Rago, and R. K. Mehra. Covari- ance intersection algorithm for distributed spacecraft state estimation. In American Control Conference , volume 6, pages 4398–4403. IEEE, 2001

    work page 2001

  8. [7]

    Bar-Shalom and L

    Y. Bar-Shalom and L. Campo. The effect of the common process noise on the two-sensor fused-track covariance. Transactions on aerospace and electronic systems, (6):803–805, 1986

    work page 1986

Show all 38 references
  1. [8]

    L. C. Carrillo-Arce, E. D. Nerurkar, Jos´ e L. Gordillo, and S. I. Roumeliotis. Decentralized multi-robot co- operative localization using covariance intersection. In International Conference on Intelligent Robots and Systems , pages 1412–1417. IEEE, 2013

    2013

  2. [9]

    L. Chen, P. O. Arambel, and R. K. Mehra. Estima- tion under unknown correlation: Covariance inter- section revisited. Transactions on Automatic Con- trol, 47(11):1879–1882, 2002

    2002

  3. [10]

    Cros, P.-O

    C. Cros, P.-O. Amblard, C. Prieur, and J.-F. Da Rocha. Split covariance intersection with corre- lated components for distributed estimation. arXiv preprint arXiv:2403.03543, 2024

    2024 arXiv

  4. [11]

    Z. Deng, P. Zhang, W. Qi, J. Liu, and Y. Gao. Se- quential covariance intersection fusion Kalman filter. Information Sciences, 189:293–309, 2012

    2012

  5. [12]

    Forsling, A

    R. Forsling, A. Hansson, F. Gustafsson, Z. Sjanic, J. L¨ ofberg, and G. Hendeby. Conservative linear unbiased estimation under partially known covari- ances. Transactions on Signal Processing , 70:3123– 3135, 2022

    2022

  6. [13]

    Franken and A

    D. Franken and A. Hupper. Improved fast covariance intersection for distributed data fusion. In 7th In- ternational Conference on Information Fusion , vol- ume 1, pages 7–pp. IEEE, 2005

    2005

  7. [14]

    Q. Guo, S. Chen, H. Leung, and S. Liu. Covari- ance intersection based image fusion technique with application to pansharpening in remote sensing. In- formation Sciences, 180(18):3434–3443, 2010

    2010

  8. [15]

    He, H.-S

    S. He, H.-S. Shin, S. Xu, and A. Tsourdos. Dis- tributed estimation over a low-cost sensor network: A review of state-of-the-art. Information Fusion , 54:21–43, 2020

    2020

  9. [16]

    R. A. Horn and C. R. Johnson. Matrix analysis . Cambridge university press, 2012

    2012

  10. [17]

    J. Hu, L. Xie, and C. Zhang. Diffusion kalman filter- ing based on covariance intersection. Transactions on Signal Processing, 60(2):891–902, 2011

    2011

  11. [18]

    S. J. Julier and J. K. Uhlmann. A non-divergent esti- mation algorithm in the presence of unknown corre- lations. In American Control Conference, volume 4, pages 2369–2373. IEEE, 1997. 20

    1997

  12. [19]

    S. J. Julier and J. K. Uhlmann. General decen- tralized data fusion with covariance intersection (ci). Handbook of Multisensor Data Fusion , 2001

    2001

  13. [20]

    S. J. Julier and J. K. Uhlmann. Simultaneous lo- calisation and map building using split covariance intersection. In International Conference on Intel- ligent Robots and Systems , volume 3, pages 1257–

  14. [21]

    S. J. Julier and J. K. Uhlmann. Using covariance intersection for SLAM. Robotics and Autonomous Systems, 55(1):3–20, 2007

    2007

  15. [22]

    W. Kahan. Circumscribing an ellipsoid about the in- tersection of two ellipsoids. Canadian Mathematical Bulletin, 11(3):437–441, 1968

    1968

  16. [23]

    S. G. Krantz and H. R. Parks. The implicit function theorem: history, theory, and applications . Springer Science & Business Media, 2002

    2002

  17. [24]

    J. Lai, Y. Zhou, J. Lin, Y. Cong, and J. Yang. Coop- erative localization based on efficient covariance in- tersection. Communications Letters, 23(5):871–874, 2019

    2019

  18. [25]

    Li and F

    H. Li and F. Nashashibi. Cooperative multi-vehicle localization using split covariance intersection fil- ter. Intelligent transportation systems magazine , 5(2):33–44, 2013

    2013

  19. [26]

    H. Li, F. Nashashibi, and M. Yang. Split covari- ance intersection filter: Theory and its application to vehicle localization. Transactions on Intelligent Transportation Systems, 14(4):1860–1871, 2013

    2013

  20. [27]

    X. R. Li, Y. Zhu, J. Wang, and C. Han. Optimal lin- ear estimation fusion. i. unified fusion rules.Transac- tions on Information Theory, 49(9):2192–2208, 2003

    2003

  21. [28]

    A. Lima, P. Bonnifait, V. Cherfaoui, and J. Al Hage. Data fusion with split covariance intersection for co- operative perception. In International Intelligent Transportation Systems Conference , pages 1112–

  22. [29]

    W. Niehsen. Information fusion based on fast co- variance intersection filtering. In 5th International Conference on Information Fusion , volume 2, pages 901–904. IEEE, 2002

    2002

  23. [30]

    Noack, J

    B. Noack, J. Sijs, and U. D. Hanebeck. Inverse co- variance intersection: New insights and properties. In 20th International Conference on Information Fu- sion, pages 1–8. IEEE, 2017

    2017

  24. [31]

    Noack, J

    B. Noack, J. Sijs, M. Reinhardt, and U. D. Hanebeck. Decentralized data fusion with inverse covariance in- tersection. Automatica, 79:35–41, 2017

    2017

  25. [32]

    Pierre, R

    C. Pierre, R. Chapuis, R. Aufr` ere, J. Laneurit, and C. Debain. Range-only based cooperative localiza- tion for mobile robots. In 21st International Confer- ence on Information Fusion, pages 1933–1939. IEEE, 2018

    1933

  26. [33]

    Reinhardt, B

    M. Reinhardt, B. Noack, P. O. Arambel, and U. D. Hanebeck. Minimum covariance bounds for the fu- sion under unknown correlations. Signal Processing Letters, 22:1210–1214, 2015

    2015

  27. [34]

    Reinhardt, B

    M. Reinhardt, B. Noack, and U. D. Hanebeck. Closed-form optimization of covariance intersection for low-dimensional matrices. In 15th International Conference on Information Fusion, pages 1891–1896. IEEE, 2012

    2012

  28. [35]

    J. K. Uhlmann. General data fusion for estimates with unknown cross covariances. In Signal Process- ing, Sensor Fusion, and Target Recognition V , vol- ume 2755, pages 536–547. SPIE, 1996

    1996

  29. [36]

    J. K. Uhlmann. Covariance consistency methods for fault-tolerant distributed data fusion. Information Fusion, 4(3):201–215, 2003

    2003

  30. [38]

    Zhang, G

    J. Zhang, G. Scebba, and W. Karlen. Covariance intersection to improve the robustness of the pho- toplethysmogram derived respiratory rate. In 42nd Annual International Conference of the IEEE Engi- neering in Medicine & Biology Society , pages 5939–

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.