Three-Dimensional Bubbly Flow Measurement Using Event-based Vision Sensor Cameras

A Al Brahim; A Taylor; B Lecordier (CORIA); K Rajamanickam; Y Hardalupas

REVIEW 3 major objections 6 minor 7 references

This paper establishes that a multi-camera event-based vision system can measure three-dimensional bubble motion and interactions to sub-millimeter synthetic accuracy, with physical velocity estimates agreeing above 97%.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-01 14:10 UTC pith:WCBAZ2TP

load-bearing objection Plausible engineering demonstration of three-camera EVS bubble tracking, but the validation is internal and the accuracy claims outrun the evidence. the 3 major comments →

arxiv 2607.18823 v1 pith:WCBAZ2TP submitted 2026-07-21 physics.class-ph

Three-Dimensional Bubbly Flow Measurement Using Event-based Vision Sensor Cameras

A Al Brahim , K Rajamanickam , B Lecordier (CORIA) , A Taylor , Y Hardalupas This is my paper

classification physics.class-ph

keywords event-based vision sensorsthree-dimensional bubble trackingbubbly flowmulti-camera reconstructionpulsed illuminationstreak velocimetrybubble-bubble interactionsparticle tracking

verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper aims to show that event-based cameras, which record only changes in light intensity, can replace high-speed frame cameras for three-dimensional bubbly-flow measurement. A three-camera array around an octagonal tank, synchronized pulsed LEDs, and a multi-view tracking pipeline produce smooth 3D trajectories for particles and bubbles, including dense bubble plumes with overlapping projections. Synthetic validation gives sub-millimeter reconstruction errors (RMSE 0.015 to 0.36 mm), and physical tests show streak-based velocity estimates agree with centroid tracking by 93.5% for particles and over 97% for bubbles. The claim matters because event cameras promise high temporal resolution, long recording, and low data rates for transient multiphase flows at lower cost than conventional high-speed cameras.

Core claim

The central discovery is that a synchronized multi-event-camera system, despite each camera lacking depth and firing only on intensity changes, can resolve three-dimensional bubble motion and bubble-bubble interactions if the illumination and reconstruction are designed together. The authors demonstrate that pulsed backlighting produces binary shadow frames that can be calibrated and triangulated like conventional images, while continuous backlighting yields streak patterns whose length encodes velocity. Combining silhouette centroid tracking, reflection-spot tracking, epipolar matching, and DLT triangulation yields continuous trajectories through dense, overlapping projections. Validation a

What carries the argument

The load-bearing pieces are: (1) a three-camera EVS array with pulsed and continuous LED illumination, where pulsed mode creates binary shadow frames for calibration and segmentation and continuous mode creates motion streaks; (2) dot-pattern calibration projections that supply intrinsic and extrinsic camera parameters; (3) a multi-view reconstruction pipeline with cross-view object-count correction, Euclidean Distance Transform splitting of merged bubble silhouettes, pairwise epipolar matching via Hungarian assignment, and SVD-based Direct Linear Transformation for triangulation; and (4) physics-constrained temporal linking with velocity-history prediction. The streak-length relation, in wh

Load-bearing premise

The load-bearing premise is that the calibration from pulsed-illumination dot patterns is unbiased; if that calibration is systematically wrong, the synthetic RMSE cannot detect it, and the physical agreement only shows internal consistency between two features of the same camera data.

What would settle it

Record a bubble plume with the three event cameras while simultaneously imaging the same volume with an independently calibrated high-speed camera or a robot-traversed reference target; if the independently measured 3D positions deviate by more than the reported RMSE, or velocity disagreement exceeds the reported few percent, the central accuracy claim would fail.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

Event-based multi-camera systems can serve as a lower-cost, longer-recording alternative to high-speed CMOS cameras for laboratory bubbly-flow measurements.
Pulsed illumination turns asynchronous event streams into binary shadow frames, so standard calibration and triangulation tools can be reused for event data.
Continuous-illumination streak lengths provide an independent per-frame velocity measurement, useful for cross-checking trajectory tracking without extra hardware.
The pipeline handles dense bubble plumes where silhouettes merge in individual views, by enforcing cross-camera epipolar consistency and splitting merged regions.
Event oversaturation in a single camera can break synchronization and spoil the whole multi-view dataset, so event-rate control is a practical constraint in ultra-dense flows.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

If the streak-velocity relation is as accurate as reported, the same technique could extend to non-bubble particles and droplets, and streak length could serve as an online self-check for frame-based trackers.
The oversaturation limitation suggests a testable adaptive scheme: dynamically reduce the region of interest or pulse frequency when event rates spike, preserving synchronization in dense regimes.
Because the pipeline resolves projection-crossing trajectories, the same mismatch-resolution logic might extend to volumetric velocimetry in turbulent multiphase flows where occlusions are frequent.
Independent trajectory ground truth, such as a robot-driven target traversing the tank, would test whether the sub-millimeter synthetic RMSE transfers to real-world geometric accuracy.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

Plausible engineering demonstration of three-camera EVS bubble tracking, but the validation is internal and the accuracy claims outrun the evidence.

read the letter

Colleague,

If you're tracking the event-camera-for-flow literature, this is a plausible engineering demonstration but its accuracy claims outrun the evidence. Three synchronized EVS cameras track rising bubble plumes in 3D, with a custom MATLAB pipeline for epipolar matching, triangulation, and temporal linking. The new-ish contribution is the application to dense bubbly flow plus a clever dual-mode check: silhouette centroid tracking vs streak-length velocity from continuous illumination, which agree to 93.5–97%. That cross-check is genuinely useful.

What the paper does well: the synthetic tests (projection-crossing, spiral, staggered releases) are a reasonable way to exercise the matching and DLT code, and the RMSE values of 0.015–0.36 mm show the algorithm is stable when fed idealized images. The experimental setup is described in enough detail to rebuild. The trajectories look smooth and physically consistent, and they don't hide the hardware issues: oversaturation, event loss during calibration (Fig. 12), and desync risk.

The soft spots are exactly the ones in the stress-test note, and they land. The synthetic renders use the same intrinsic/extrinsic parameters the tracker consumes, so a biased calibration is invisible in those RMSEs. The physical velocity agreement compares two quantities reconstructed with the same projection matrices; a scale error cancels in the ratio, so 97% consistency is compatible with a 10% or larger absolute error. There is no independent ground truth: no high-speed camera, no translation stage, no known-velocity object. The high-density 45 Hz case is only described qualitatively, so the quantitative validation is confined to the low-density plume. Also, the abstract promises 'morphology' measurements but the results section never presents any. No code or data is provided, which is limiting but not disqualifying for a conference paper.

None of this kills the central idea; a multi-EVS system for bubble plumes is plausible and the internal consistency is encouraging. But the sub-millimeter accuracy claim should be read as algorithmic noise, not real-world calibration accuracy. The paper needs an independent calibration check (a target on a moved stage, or simultaneous HS camera) before the external claims are credible.

Worth sending to peer review — it's a serious application with enough detail to referee. I'd ask the authors to either add that independent validation or soften the accuracy claims, and to fix the morphology mismatch. For me, I'd cite it cautiously, only as a demonstration of concept.

Referee Report

3 major / 6 minor

Summary. The paper describes a three-camera Event-Based Vision Sensor (EVS) system for three-dimensional tracking of particles and bubbles in a water tank. The authors develop a custom processing pipeline that segments event frames, performs epipolar matching, triangulates via DLT, and links centroids into trajectories with physics-constrained tracking. Validation is two-fold: synthetic renderings of analytically prescribed trajectories (RMSE 0.015–0.36 mm) and physical experiments with particle releases and bubble plumes, where velocities from centroid tracking and event-streak analysis agree to 93.5–97%. The paper also discusses event oversaturation as a hardware limitation.

Significance. The paper addresses a timely and important problem—high-speed multiphase flow measurement—using event cameras, which offer low data rates and high temporal resolution. The proposed multi-camera framework is a plausible alternative to high-speed CMOS systems, and the synthetic benchmark provides a useful check of the algorithmic pipeline. However, the headline accuracy claims (sub-millimeter RMSE, >97% velocity consistency) are not supported by independent ground truth for the physical experiments. If the authors add such validation or substantially temper their claims, the methodology could be a valuable contribution to multiphase flow diagnostics.

major comments (3)

[Section 3 (Image analysis and tracking algorithms), synthetic validation paragraph] The synthetic rendering uses 'the same intrinsic and extrinsic calibration parameters obtained from the experimental calibration procedure.' Hence the reported RMSE values (0.224, 0.362, 0.015 mm) measure only algorithmic consistency; any systematic bias in calibration (e.g., from event loss during calibration, Fig. 12) is common to both the rendering and the reconstruction and does not appear in the error. The Abstract and Conclusions therefore overstate the claim of 'sub-millimeter global accuracy' for physical measurements. Please either qualify the claim as algorithmic precision or add a validation against independent geometric ground truth (e.g., a target with known 3D coordinates or a mechanical traverse).
[Section 4 (Results and Discussion), velocity comparison paragraph] The 93.5–97% velocity agreement between centroid tracking and streak-length estimates is an internal cross-check: both methods use the same three-camera event data and the same DLT projection matrices. A scale error in the calibration (e.g., biased baseline or focal length) multiplies both reconstructed velocities by the same factor, so the velocity ratio is insensitive to absolute calibration error. No independent reference (high-speed camera, known-velocity object, mechanical stage) is provided. Thus the physical experiments demonstrate consistency but not absolute accuracy. The sub-millimeter accuracy claim cannot be extrapolated to the experimental results.
[Abstract and Conclusions] The Abstract states 'sub-millimeter global accuracy' without specifying that this is only for the synthetic benchmark, and the Conclusions mention 'high-frequency (16–60 Hz)' bubble plumes although the paper only reports experiments at 16 Hz and 45 Hz (10 and 20 mL/min). Also, the claim of resolving 'dense bubble-bubble interactions' is supported only by example trajectories; no quantitative metric (e.g., tracking success rate, maximum resolvable density, or uncertainty) is provided. Please qualify the accuracy claim, correct the frequency range, and consider adding quantitative measures of tracking robustness.

minor comments (6)

[Section 3, Eq. (1)] Equation (1) is garbled in the manuscript ('?㌵?㌵...'). The relationship between streak length, accumulation time, and velocity should be typeset properly and the variables defined explicitly.
[Section 3, text near Figure 4] There is a typo: 'Cusing DaVis' should read 'Using DaVis.'
[Section 2, calibration paragraph] The paper mentions event loss during calibration target acquisition (Fig. 12) but does not state whether the calibration sequence was re-recorded after the oversaturation event or whether any corrupted frames were discarded. Please clarify how the calibration dataset was sanitized.
[Section 3, synthetic cases] The synthetic validation consists of only three cases; please specify the number of particles/bubbles and trajectories in each case, and discuss whether the RMSE values are representative of the full parameter space.
[Section 3, segmentation] The 'median bubble size' used for morphological decomposition is not defined quantitatively. It would be helpful to state how it is obtained (e.g., from the entire sequence) and how sensitive the decomposition is to this parameter.
[Section 3, tracking parameters] The 'maximum physically admissible displacement per frame' and 'maximum allowable frame gap' are free parameters. Please provide the chosen values and justify them with respect to the expected bubble velocities and frame rates.

Circularity Check

0 steps flagged

No significant circularity: the validation is self-referential with respect to calibration, but no central result reduces to its inputs by construction.

full rationale

The paper's central deliverable is a measurement framework, not a derived law, and its quantitative claims are internal consistency checks plus a synthetic inversion test. The synthetic benchmark renders ground-truth trajectories "using the same intrinsic and extrinsic calibration parameters obtained from the experimental calibration procedure" (Section 3), so it cannot detect systematic calibration bias; this is an external-validity limitation, not a circular derivation, because the ground-truth 3D trajectories are analytically prescribed and the RMSE (0.015-0.362 mm) is an unforced measure of the DLT/epipolar/temporal-association pipeline's inversion error. The physical velocity comparison between centroid tracking and streak-length estimates shares the same event streams and calibration, and the paper calls the streak estimate "independent" only in the sense of a different image feature; the 93.5-97% agreement is an empirical cross-check, not an identity by construction. The only self-citation, "our previous work (Al Brahim et al., n.d.)", is background motivation and is not load-bearing; there is no imported uniqueness theorem and no fitted parameter renamed as a prediction. The paper's own note that oversaturation during calibration acquisition "could have severely compromised the calibration dataset" (Fig. 12) is a correctness risk, not a circular step. Therefore no specific reduction from output to input can be exhibited, and the circularity score is 0.

Axiom & Free-Parameter Ledger

5 free parameters · 4 axioms · 0 invented entities

No new physical entities are introduced. The central claim rests on standard camera-calibration/triangulation assumptions plus several unspecified algorithm thresholds. The key epistemic gap is that the synthetic benchmark reuses the same calibration parameters, so calibration error is not independently bounded.

free parameters (5)

Camera calibration parameters (intrinsic/extrinsic) = Not reported numerically
Fitted to pulsed-illumination dot-target images in DaVis; used for all DLT triangulation. Any bias is invisible to the synthetic validation.
Maximum admissible displacement per frame = Not stated
Used in Lagrangian nearest-neighbour linking; value derived from an unstated upper-bound velocity, affects trajectory association.
Maximum allowable frame gap = Not stated
Temporal coherence constraint for re-association after detection loss; not quantified.
Median bubble size for morphological decomposition = Computed from data
Splitting merged regions uses median bubble size across all frames; an ad hoc, data-dependent threshold.
Gaussian smoothing sigma for EDT distance maps = Not stated
Gaussian filter applied to distance maps before local-maxima detection; no sigma given.

axioms (4)

standard math Pinhole camera model with DLT/SVD yields unbiased 3D positions
Section 3 uses DLT formulation; requires camera calibration to be correct.
domain assumption Pulsed illumination produces binary frames whose silhouettes faithfully represent bubble shadows
Section 3 relies on reconstructed frames resembling binary HS images; event loss and noise can corrupt silhouettes.
domain assumption Streak length under continuous illumination equals velocity times accumulation time
Equation 1 defines streak-based velocity; assumes clean interface detection and constant LED illumination.
ad hoc to paper Cross-view maximum object count and median-size-based splitting correctly resolve merged detections
Section 3 uses this heuristic to handle overlaps; no validation of the splitting against known bubble counts.

pith-pipeline@v1.3.0-alltime-deepseek · 10504 in / 10489 out tokens · 88515 ms · 2026-08-01T14:10:57.539224+00:00 · methodology

0 comments

read the original abstract

A three-camera Event-Based Vision Sensor (EVS) system is employed to perform three-dimensional measurements of bubble motion, morphology, and bubble{\DH}bubble interactions. The multi-EVS configuration mitigates the absence of direct depth information inherent to binary event-based imaging while preserving key advantages, including high temporal resolution, low latency, and reduced data throughput. The experimental configuration consisted of an octagonal tank equipped with a controlled particle release mechanism and an air diffuser. Camera synchronization and pulsed LED illumination were achieved using a dedicated signal generator and driver circuitry, while calibration was performed using pulsed-illumination recordings of a target acquired at multiple depths. A comprehensive, inhouse computational framework was developed to process the event data for three-dimensional motion trajectory reconstruction. The validation of the developed tracking framework followed a rigorous multi-stage pipeline to ensure reconstruction fidelity. The framework was initially benchmarked against synthetic rendering cases of increasing kinematic complexity to evaluate 3D trajectory reconstruction accuracy under controlled conditions, achieving sub-millimeter global accuracy with root-mean-square error (RMSE) values ranging from 0.015 to 0.36 mm. Following numerical validation, physical baseline experiments were conducted using precisely manufactured particle releases through both a gated chamber and a single-particle claw opening mechanism. Subsequently, dynamic bubble plumes generated via multiple inlets across various compressed air flow rates were evaluated. The EVS framework successfully resolved dense bubble-bubble interactions, producing smooth, physically consistent three-dimensional trajectories across all tested conditions. Quantitative results showed strong agreement, yielding an overall velocity consistency exceeding 97% between conventional centroid tracking and independent velocity estimates derived from continuous-illumination event streaks. Overall, this methodology demonstrates a robust framework for high-speed volumetric tracking and bubble flow measurement. However, event oversaturation remains a key hardware limitation in ultra-dense regimes, as oversaturation in a single camera can cause system desynchronization.

discussion (0)

Reference graph

Works this paper leans on

7 extracted references · 5 canonical work pages

[1]

I., & Karara, H

Abdel-Aziz, Y. I., & Karara, H. M. (2015). Direct Linear Transformation from Comparator Coordinates into Object Space Coordinates in Close-Range Photogrammetry. Photogrammetric Engineering & Remote Sensing, 81(2), 103–107. https://doi.org/10.14358/PERS.81.2.103 Al Brahim, A., Rajamanickam, K., Taylor, A. M. K. P., & Hardalupas, Y. (n.d.). Assessment of Ev...

Pith/arXiv arXiv 2015
[6]

E., & Klinner, J

https://doi.org/10.1007/s00348-023-03641-8 22nd LISBON Laser Symposium 2026 Willert, C. E., & Klinner, J. (2025). Dynamic wall shear stress measurement using event-based 3d particle tracking. Experiments in Fluids, 66(2),

doi:10.1007/s00348-023-03641-8 2026
[32]

https://doi.org/10.1007/s00348-024-03946-2

work page doi:10.1007/s00348-024-03946-2
[33]

https://doi.org/10.1063/5.0057198 Tagawa, Y., Takagi, S., & Matsumoto, Y. (2014). Surfactant effect on path instability of a rising bubble. Journal of Fluid Mechanics, 738, 124–142. https://doi.org/10.1017/jfm.2013.571 Tan, S., Zhong, S., & Ni, R. (2023). 3D Lagrangian tracking of polydispersed bubbles at high image densities. Experiments in Fluids, 64(4),

work page doi:10.1063/5.0057198 2014
[34]

https://doi.org/10.1007/s00348-025-03963-9 Wang, Y., Idoughi, R., & Heidrich, W. (2020). Stereo Event-Based Particle Tracking Velocimetry for 3D Fluid Flow Reconstruction. In A. Vedaldi, H. Bischof, T. Brox, & J.-M. Frahm (Eds.), Computer Vision – ECCV 2020 (pp. 36–53). Springer International Publishing. https://doi.org/10.1007/978-3-030- 58526-6_3 Willer...

work page doi:10.1007/s00348-025-03963-9 2020
[85]

https://doi.org/10.1007/s00348-023-03601-2 Wang, H., Xu, Y., & Wang, J. (2025). Experimental study on bubble pairs and induced flow fields using tomographic particle image velocimetry. Experiments in Fluids, 66(5),

work page doi:10.1007/s00348-023-03601-2 2025
[98]

https://doi.org/10.1007/s00348-025-04026-9 Wang, L., Ma, T., Lucas, D., Eckert, K., & Hessenkemper, H. (2025). A contribution to 3D tracking of deformable bubbles in swarms using temporal information. Experiments in Fluids, 66(2),

work page doi:10.1007/s00348-025-04026-9 2025

[1] [1]

I., & Karara, H

Abdel-Aziz, Y. I., & Karara, H. M. (2015). Direct Linear Transformation from Comparator Coordinates into Object Space Coordinates in Close-Range Photogrammetry. Photogrammetric Engineering & Remote Sensing, 81(2), 103–107. https://doi.org/10.14358/PERS.81.2.103 Al Brahim, A., Rajamanickam, K., Taylor, A. M. K. P., & Hardalupas, Y. (n.d.). Assessment of Ev...

Pith/arXiv arXiv 2015

[2] [6]

E., & Klinner, J

https://doi.org/10.1007/s00348-023-03641-8 22nd LISBON Laser Symposium 2026 Willert, C. E., & Klinner, J. (2025). Dynamic wall shear stress measurement using event-based 3d particle tracking. Experiments in Fluids, 66(2),

doi:10.1007/s00348-023-03641-8 2026

[3] [32]

https://doi.org/10.1007/s00348-024-03946-2

work page doi:10.1007/s00348-024-03946-2

[4] [33]

https://doi.org/10.1063/5.0057198 Tagawa, Y., Takagi, S., & Matsumoto, Y. (2014). Surfactant effect on path instability of a rising bubble. Journal of Fluid Mechanics, 738, 124–142. https://doi.org/10.1017/jfm.2013.571 Tan, S., Zhong, S., & Ni, R. (2023). 3D Lagrangian tracking of polydispersed bubbles at high image densities. Experiments in Fluids, 64(4),

work page doi:10.1063/5.0057198 2014

[5] [34]

https://doi.org/10.1007/s00348-025-03963-9 Wang, Y., Idoughi, R., & Heidrich, W. (2020). Stereo Event-Based Particle Tracking Velocimetry for 3D Fluid Flow Reconstruction. In A. Vedaldi, H. Bischof, T. Brox, & J.-M. Frahm (Eds.), Computer Vision – ECCV 2020 (pp. 36–53). Springer International Publishing. https://doi.org/10.1007/978-3-030- 58526-6_3 Willer...

work page doi:10.1007/s00348-025-03963-9 2020

[6] [85]

https://doi.org/10.1007/s00348-023-03601-2 Wang, H., Xu, Y., & Wang, J. (2025). Experimental study on bubble pairs and induced flow fields using tomographic particle image velocimetry. Experiments in Fluids, 66(5),

work page doi:10.1007/s00348-023-03601-2 2025

[7] [98]

https://doi.org/10.1007/s00348-025-04026-9 Wang, L., Ma, T., Lucas, D., Eckert, K., & Hessenkemper, H. (2025). A contribution to 3D tracking of deformable bubbles in swarms using temporal information. Experiments in Fluids, 66(2),

work page doi:10.1007/s00348-025-04026-9 2025