Pith. sign in

REVIEW 4 major objections 6 minor 68 references

Motion Segmentation and Egomotion Estimation from Event-Based Normal Flow

T0 review · 4 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read Event-camera normal flow, not full optical flow, can drive motion segmentation and egomotion estimation.

desk verdict Novel pipeline, but translation-scale ambiguity and thin evaluation keep it from proving its headline claim. read the letter →

arxiv 2507.14500 v1 pith:B4OR3VWX submitted 2025-07-19 cs.CV cs.RO

Zhiyuan Hua , Dehao Yuan , Cornelia Fermüller This is my paper · ORCID
classification cs.CVcs.RO
keywords event-basedvisionnormalflowmotionsegmentationegomotionestimationneuromorphicsensorsEVIMO2v2hierarchicalclusteringvisual-inertial
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 is trying to establish that motion segmentation and egomotion estimation can be solved from event-based normal flow, the one-dimensional component of optical flow along the image gradient, rather than from full optical flow or depth. Its pipeline over-segments each event slice, fits a planar rigid-motion model to the background, and uses the residuals as evidence of independently moving objects, then merges segments by temporal consistency and motion similarity. Using IMU rotation to remove rotation, it estimates the translational motion of the camera and of each object. On EVIMO2v2, the reported IoU ranges from 73.36% to 82.15% across five scenes and per-axis translation RMSE stays below 0.1 m/s, which is the quantitative basis for the claim that full optical flow is unnecessary. A sympathetic read is that this gives event-based robots a cheaper path to scene motion understanding than optical-flow pipelines.

What carries the argument

The load-bearing object is the normal-flow constraint equation $u_n(x) = (C(x)n_0)^\top a$, which expresses the normal flow at a pixel as a linear function of an eight-parameter vector $a$ combining the camera's rotational and translational motion with the coefficients of an assumed planar scene. Fitting this model by least squares and clustering the residuals yields the initial foreground/background mask; the same constraint is re-solved on identified background pixels, using IMU rotation to derotate the normal flow, to estimate translation. Segmentation is carried by a recursive pipeline: k-means over-segmentation of events in space and normal-flow space, temporal-consistency warping of the previous background mask, and hierarchical merging of clusters whose fitted translations and residuals are similar, with a penalty preventing background/foreground merges.

What would settle it

Run the pipeline on an EVIMO2v2 or similar indoor sequence in which the camera moves but no independently moving object is present, and record the foreground mask produced by the initial residual clustering; if a substantial fraction of static, depth-discontinuous background is labeled as moving, then the single-plane assumption is the point of failure. A more direct check is to compare segmentation IoU on sequences with planar backgrounds versus heavily cluttered, non-planar backgrounds.

Watch

Extended reading notes

Core claim

The central claim is that the one-dimensional normal flow measurable at each event, combined with IMU rotation measurements, is sufficient to jointly segment independently moving objects and estimate camera and object translation. The paper derives a linear constraint linking normal flow to the camera's 3D translation and to a planar scene model, solves it by least squares, and treats large residuals as evidence of independent motion. It then refines the coarse foreground/background separation through a recursive pipeline that warps previous masks, merges over-segmented clusters by temporal consistency, and applies hierarchical clustering with a motion-similarity score. The reported experiments on EVIMO2v2 show that this normal-flow-only formulation reaches the IoU and RMSE levels in Tables 1 and 2, including accurate behavior at object boundaries where full optical flow methods typically struggle.

Load-bearing premise

The initial foreground/background split assumes the whole scene is a single plane, so large residuals from the planar fit are read as independent motion; in an indoor room, static objects at different depths also create large residuals, which would contaminate the initial mask if the plane assumption fails.

Editorial extensions

If this is right

  • Event-driven robots can detect and track independent objects in real time because the pipeline consumes only local normal flow and IMU rotation, not dense optical flow.
  • The residual analysis provides a confidence signal: regions with persistently high residuals after background fitting are exactly where new independent moving objects should be searched.
  • The recursive design maintains consistent track IDs for background and foreground over time, which supports downstream tasks like collision avoidance and object interaction.
  • On EVIMO2v2 the segmentation IoU stays in the mid-70s to low-80s across scenes, indicating the formulation transfers across different room geometries without per-scene retraining of the optimization stages.

Reading between the lines

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

  • A testable extension is to replace the single-plane initial fit with a multi-plane or inverse-depth model; the comparison would show how much of the segmentation error is inherited from the planar assumption.
  • The motion-similarity merging rule could be lifted to stereo event streams, where the two-view constraint might resolve the depth-scale ambiguity that keeps the monocular translation estimate partial.
  • Benchmarking this pipeline with a classical plane-fitting normal-flow estimator instead of the learned one would isolate how much accuracy comes from the normal-flow quality rather than the segmentation design.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper proposes an iterative, optimization-based pipeline for motion segmentation and translational motion estimation from event-based normal flow and IMU rotation. It over-segments events by clustering normal flow and image coordinates, fits a planar motion model to segregate foreground from background via residual analysis, merges clusters using temporal consistency and motion similarity, and estimates per-cluster translation with depth set to Z=1. Experiments on EVIMO2v2 report IoU for five sequences and per-axis velocity RMSE for three sequences.

Significance. If the results hold, the contribution is interesting because it tackles joint segmentation and egomotion from normal flow alone, avoiding full optical flow and explicit depth estimation, and it builds directly on a recent learned normal-flow estimator. The geometric formulation in Sec. 3.3 is standard and the recursive use of temporal consistency is sensible. However, the quantitative claims, especially meter-per-second velocity accuracy, are not currently supported by the evaluation as written.

major comments (4)
  1. [Sec. 3.5, Eq. (3), Table 2] The reported translational velocity RMSE in meters per second is not reproducible from the described method because normal flow is scale-ambiguous. Equation (3) depends on t and Z only through the ratio t/Z, so a global scaling (t, Z) -> (s t, s Z) leaves the measurement unchanged. The IMU supplies only rotation, the planar fit in Sec. 3.3 is scale-ambiguous, and Sec. 3.5 explicitly sets Z=1 when solving for cluster translations, which fixes an arbitrary gauge but not metric scale. No step in the pipeline recovers absolute scale. If a scale source is used (e.g., known plane distance, stereo depth, or ground-truth pose), it must be disclosed; otherwise the Table 2 values are not interpretable and the abstract's claim of translational motion estimation requires revision to up-to-scale or direction-only estimation.
  2. [Sec. 3.3] The preliminary foreground-background segregation assumes the scene is a single plane and treats residuals from that planar fit as evidence of independent motion. In an indoor dataset such as EVIMO2v2, static background structures at different depths also violate the planar assumption and can produce large residuals, contaminating the initial background mask that all later temporal-consistency and motion-similarity stages inherit. The paper reports no experiment isolating this failure mode; a static-scene residual analysis or an ablation with nonplanar backgrounds is needed to support the robustness claim.
  3. [Section 4, Tables 1 and 2] The evaluation contains no comparison with any existing motion-segmentation or egomotion method, despite claims of 'accurate segmentation' and 'significant advantages at object boundaries.' Reporting IoU and RMSE in isolation does not substantiate those claims. The authors should compare against at least one event-based baseline (e.g., contrast maximization [57], normal-flow-based motion/structure estimation [53], or a learning-based motion segmentation method) and report per-sequence variability or error bars, as the current single-number results do not support the strength of the stated conclusions.
  4. [Sec. 4.2] The object translation evaluation explicitly drops the depth component and does not account for object rotation, evaluating only image-plane ΔX and ΔY. This is a stated limitation, but it directly narrows the abstract's unqualified claim of 'translational motion estimation.' The camera egomotion table includes Vz, yet the object evaluation excludes the analogous component; the discrepancy should be acknowledged in the abstract or the claims should be limited to image-plane translation.
minor comments (6)
  1. [Sec. 2.4] There is a typo in the reference citation: 'Yuan et al. [62]]' contains a double closing bracket.
  2. [Fig. 3, Table 1] Figure 3 includes per-frame IoU for sequence 'box-00' from EVIMO1, but Table 1 only lists EVIMO2v2 scenes; clarify the quantitative status of the EVIMO1 sequence and the criterion for including it.
  3. [Sec. 3.4] The 'linear SVM' formulation for estimating background translation is not explained; either derive it or provide a precise reference to the formulation in [62].
  4. [Secs. 3.2--3.5] Several hyperparameters (cluster count 30, λ=0.5, λr=0.5, Tikhonov λ=1e-6, merging threshold, EMA adaptive α, Gaussian kernel size) are given without sensitivity analysis or a principled choice; a robustness check would strengthen the method's reproducibility.
  5. [Eq. (8)] The entries of the vector a in Eq. (8) are not individually labeled; adding row labels or a short explanation would improve readability.
  6. [Table 2] The RMSE table does not report the number of evaluated frames or slices; adding this information is necessary for the reader to judge the statistical reliability of the errors.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the pipeline derives segmentation and motion from normal flow and IMU inputs, and its self-citations provide external components rather than fitted targets.

full rationale

The paper's derivation chain is self-contained with respect to its stated inputs: per-event normal flow, IMU rotation, and event coordinates. The geometric model in Eqs. (2)-(8) relates normal flow to 3D motion and planar scene structure; residuals from a least-squares planar fit seed a coarse foreground/background separation; temporal warping and hierarchical motion-similarity merging refine the segmentation. None of these stages fits a parameter to the ground-truth segmentation masks or to the reported velocity values. The main self-references are citations [61,62] for a learned normal-flow estimator and an SVM-based translation solver; these are external, previously published components that the paper consumes as inputs, not quantities fitted to this paper's target outputs. The apparent scale ambiguity—Sec. 3.5 sets Z=1, and normal flow alone cannot determine metric translation scale—is a legitimate concern about whether Table 2's m/s RMSE values are interpretable, but it is an evaluation/interpretability issue rather than a circular reduction: the reported values are not defined to equal the inputs by construction. For these reasons, no specific circular step can be exhibited, and the appropriate finding is no significant circularity.

Assumptions & free parameters 8 free parameters · 6 assumptions · 0 invented entities

The method rests on a planar-scene assumption, a learned normal flow input from the same group, a constant-depth simplification, and several hand-set thresholds. These are reasonable engineering assumptions individually, but none is validated independently in this paper.

free parameters (8)
  • Over-segmentation cluster count = 30
    Sec. 3.2 sets the number of k-means clusters to 30 to ensure over-segmentation; no sensitivity analysis is given.
  • Feature weighting factor lambda = 0.5
    Sec. 3.2 uses lambda=0.5 in Eq. (1) to balance spatial and normal-flow information.
  • Residual similarity weight lambda_r = 0.5
    Eq. (9) in Sec. 3.5 uses lambda_r=0.5 to balance translation similarity and residual coherence.
  • Tikhonov regularization lambda = 1e-6
    Sec. 3.5 uses lambda=1e-6 to regularize per-cluster translation estimation.
  • Merging similarity threshold
    Sec. 3.5 merges pairs above a similarity threshold, but the threshold value is not reported.
  • EMA adaptive alpha
    Sec. 3.4 updates the persistent background map with an adaptive alpha based on similarity; the update rule is not specified.
  • Background merge penalty strength
    Sec. 3.5 adds a penalty when only one cluster matches the persistent background map; the magnitude is not given.
  • Gaussian smoothing kernel size/sigma
    Sec. 3.3 smooths residuals with a Gaussian filter in the image grid; kernel parameters are not specified.
assumptions (6)
  • domain assumption The scene background can be modeled as a single plane for initial motion fitting.
    Sec. 3.3 fits eight plane parameters to all events and treats large residuals as independently moving objects; non-planar static structure will produce residuals.
  • domain assumption The pre-trained normal flow estimator [61,62] is accurate enough, especially at object boundaries, for the pipeline's inputs.
    The entire method consumes per-event normal flow from this learned estimator; if it fails, all downstream equations inherit the error.
  • ad hoc to paper Translation can be estimated per cluster with depth set to Z=1, making cluster translations comparable.
    Sec. 3.5 solves for t with Z=1; this equates depth-scaled translations across clusters and is not justified by any scene assumption.
  • domain assumption IMU rotation measurements are accurate and temporally aligned with event slices.
    Rotation is used to derotate normal flow in Secs. 3.3 and 3.4; rotation error propagates directly into residual analysis.
  • domain assumption Only spatially connected clusters may be merged.
    Sec. 3.5 restricts merging to clusters with overlapping bounding boxes; this design choice prevents fusing disjoint objects but also prevents merging occluded parts of the same object.
  • domain assumption A constant-depth planar model suffices for warping the previous background mask.
    Sec. 3.4 warps previous background events using prior motion and a planar model with constant depth.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Motion Segmentation and Egomotion Estimation from Event-Based Normal Flow." pith.science (2026). https://pith.science/paper/B4OR3VWX

@misc{pith2026250714500,
  author       = {Pith},
  title        = {Pith review of: Motion Segmentation and Egomotion Estimation from Event-Based Normal Flow},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/B4OR3VWX}},
  note         = {Machine review of arXiv:2507.14500}
}
read the original abstract

This paper introduces a robust framework for motion segmentation and egomotion estimation using event-based normal flow, tailored specifically for neuromorphic vision sensors. In contrast to traditional methods that rely heavily on optical flow or explicit depth estimation, our approach exploits the sparse, high-temporal-resolution event data and incorporates geometric constraints between normal flow, scene structure, and inertial measurements. The proposed optimization-based pipeline iteratively performs event over-segmentation, isolates independently moving objects via residual analysis, and refines segmentations using hierarchical clustering informed by motion similarity and temporal consistency. Experimental results on the EVIMO2v2 dataset validate that our method achieves accurate segmentation and translational motion estimation without requiring full optical flow computation. This approach demonstrates significant advantages at object boundaries and offers considerable potential for scalable, real-time robotic and navigation applications.

Figures

Figures reproduced from arXiv: 2507.14500 by the authors.

Figure 1
Figure 1. Overview of our motion segmentation pipeline. [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Visualization of the over-segmentation results. Although [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Qualitative segmentation results with per-frame IoU val [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

68 extracted references · 64 canonical work pages

  1. [57]

    Event-based motion segmentation by motion compensation

    Timo Stoffregen, Guillermo Gallego, Tom Drummond, Lindsay Kleeman, and Davide Scaramuzza. Event-based motion segmentation by motion compensation. In Proceed- ings of the IEEE/CVF International Conference on Com- puter Vision, pages 7244–7253, 2019. 2

    work page 2019

  2. [53]

    Motion and structure from event-based normal flow

    Zhongyang Ren, Bangyan Liao, Delei Kong, Jinghang Li, Peidong Liu, Laurent Kneip, Guillermo Gallego, and Yi Zhou. Motion and structure from event-based normal flow. In European Conference on Computer Vision , pages 108–

    work page

  3. [1]

    Estimat- ing visual motion using an event-based artificial retina

    Luma Issa Abdul-Kreem and Heiko Neumann. Estimat- ing visual motion using an event-based artificial retina. In Computer Vision, Imaging and Computer Graphics Theory and Applications: 10th International Joint Conference, VISI- GRAPP 2015, Berlin, Germany, March 11–14, 2015, Re- vised Selected Papers 10, pages 396–415, 2016. 2

    work page 2015

  4. [2]

    Neuromorphic vision-based motion segmentation with graph transformer neural network

    Yusra Alkendi, Rana Azzam, Sajid Javed, Lakmal Senevi- ratne, and Yahya Zweiri. Neuromorphic vision-based motion segmentation with graph transformer neural network. IEEE Transactions on Multimedia, 2024. 2

    work page 2024

  5. [3]

    Motion Segmentation for Neuromorphic Aerial Surveillance

    Sami Arja, Alexandre Marcireau, Saeed Afshar, Bharath Ramesh, and Gregory Cohen. Motion segmentation for neuromorphic aerial surveillance. arXiv preprint arXiv:2405.15209, 2024. 2

    work page Pith review arXiv 2024

  6. [4]

    Joint direct estimation of 3d geometry and 3d motion using spatio temporal gradients

    Francisco Barranco, Cornelia Ferm ¨uller, Yiannis Aloimonos, and Eduardo Ros. Joint direct estimation of 3d geometry and 3d motion using spatio temporal gradients. Pattern Recogni- tion, 113:107759, 2021. 1

    work page 2021

  7. [5]

    Event-based visual flow

    Ryad Benosman, Charles Clercq, Xavier Lagorce, Sio-Hoi Ieng, and Chiara Bartolozzi. Event-based visual flow. IEEE Transactions on Neural Networks and Learning Systems, 25 (2):407–417, 2013. 1

    work page 2013

  8. [6]

    Structure from motion: Beyond the epipolar con- straint

    Tom ´aˇs Brodsk ´y, Cornelia Ferm ¨uller, and Yiannis Aloi- monos. Structure from motion: Beyond the epipolar con- straint. International Journal of Computer Vision , 37:231– 258, 2000. 1

    work page 2000

Show all 68 references
  1. [7]

    Event-based, direct camera tracking from a photometric 3d map using nonlinear optimization

    Samuel Bryner, Guillermo Gallego, Henri Rebecq, and Da- vide Scaramuzza. Event-based, direct camera tracking from a photometric 3d map using nonlinear optimization. In 2019 International Conference on Robotics and Automation (ICRA), pages 325–331, 2019. 2

    2019

  2. [8]

    Evimo2: An event camera dataset for motion segmentation, optical flow, structure from motion, and visual inertial odometry in indoor scenes with monoc- ular or stereo algorithms

    Levi Burner, Anton Mitrokhin, Cornelia Ferm ¨uller, and Yiannis Aloimonos. Evimo2: An event camera dataset for motion segmentation, optical flow, structure from motion, and visual inertial odometry in indoor scenes with monoc- ular or stereo algorithms. arXiv preprint arXiv:22...

    arXiv

  3. [9]

    Low-latency localization by active led markers tracking using a dynamic vision sensor

    Andrea Censi, Jonas Strubel, Christian Brandli, Tobi Del- bruck, and Davide Scaramuzza. Low-latency localization by active led markers tracking using a dynamic vision sensor. In 2013 IEEE/RSJ International Conference on Intelligent Robots and Systems, pages 891–898, 2013. 2

    2013

  4. [10]

    Su- pereio: Self-supervised event feature learning for event iner- tial odometry

    Peiyu Chen, Fuling Lin, Weipeng Guan, and Peng Lu. Su- pereio: Self-supervised event feature learning for event iner- tial odometry. arXiv preprint arXiv:2503.22963, 2025. 2

    2025 arXiv

  5. [11]

    On-board real-time optic-flow for miniature event-based vision sensors

    J ¨org Conradt. On-board real-time optic-flow for miniature event-based vision sensors. In2015 IEEE International Con- ference on Robotics and Biomimetics (ROBIO), pages 1858– 1863, 2015. 2

    2015

  6. [12]

    Robotic goalie with 3 ms reaction time at 4% cpu load using event-based dynamic vi- sion sensor

    Tobi Delbruck and Manuel Lang. Robotic goalie with 3 ms reaction time at 4% cpu load using event-based dynamic vi- sion sensor. Frontiers in Neuroscience, 7:223, 2013. 2

    2013

  7. [13]

    Fast sensory mo- tor control based on event-based hybrid neuromorphic- procedural system

    Tobi Delbruck and Patrick Lichtsteiner. Fast sensory mo- tor control based on event-based hybrid neuromorphic- procedural system. In 2007 IEEE International Symposium on Circuits and Systems (ISCAS), pages 845–848, 2007. 2

    2007

  8. [14]

    Passive navigation as a pattern recogni- tion problem

    Cornelia Ferm ¨uller. Passive navigation as a pattern recogni- tion problem. International Journal of Computer Vision, 14 (2):147–158, 1995. 1

    1995

  9. [15]

    Navigational preliminaries

    Cornelia Ferm ¨uller. Navigational preliminaries. In Active Perception, pages 103–150. 2013. 1

    2013

  10. [16]

    Ambiguity in structure from motion: Sphere versus plane

    Cornelia Ferm ¨uller and Yiannis Aloimonos. Ambiguity in structure from motion: Sphere versus plane. International Journal of Computer Vision, 28:137–154, 1998. 1

    1998

  11. [17]

    Observability of 3d motion

    Cornelia Ferm ¨uller and Yiannis Aloimonos. Observability of 3d motion. International Journal of Computer Vision, 37 (1):43–63, 2000. 1

    2000

  12. [18]

    Accurate angu- lar velocity estimation with an event camera

    Guillermo Gallego and Davide Scaramuzza. Accurate angu- lar velocity estimation with an event camera. IEEE Robotics and Automation Letters, 2(2):632–639, 2017. 2

    2017

  13. [19]

    A unifying contrast maximization framework for event cam- eras, with applications to motion, depth, and optical flow es- timation

    Guillermo Gallego, Henri Rebecq, and Davide Scaramuzza. A unifying contrast maximization framework for event cam- eras, with applications to motion, depth, and optical flow es- timation. In Proceedings of the IEEE Conference on Com- puter Vision and Pattern Recognition , pages...

    2018

  14. [20]

    Focus is all you need: Loss functions for event- based vision

    Guillermo Gallego, Mathias Gehrig, and Davide Scara- muzza. Focus is all you need: Loss functions for event- based vision. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 12280– 12289, 2019. 2

    2019

  15. [21]

    Davison, J ¨org Conradt, Kostas Daniilidis, et al

    Guillermo Gallego, Tobi Delbr ¨uck, Garrick Orchard, Chiara Bartolozzi, Brian Taba, Andrea Censi, Stefan Leutenegger, Andrew J. Davison, J ¨org Conradt, Kostas Daniilidis, et al. Event-based vision: A survey. IEEE Transactions on Pattern Analysis and Machine Intelligence, 44(1...

    2020

  16. [22]

    Out of the room: Generalizing event-based dynamic motion segmenta- tion for complex scenes

    Stamatios Georgoulis, Weining Ren, Alfredo Bochicchio, Daniel Eckert, Yuanyou Li, and Abel Gawel. Out of the room: Generalizing event-based dynamic motion segmenta- tion for complex scenes. In 2024 International Conference on 3D Vision (3DV), pages 442–452, 2024. 2

    2024

  17. [23]

    Deio: Deep event inertial odometry

    Weipeng Guan, Fuling Lin, Peiyu Chen, and Peng Lu. Deio: Deep event inertial odometry. arXiv preprint arXiv:2411.03928, 2024. 2

    2024 arXiv

  18. [24]

    Pijnacker Hordijk, Kirk Y .W

    Bas J. Pijnacker Hordijk, Kirk Y .W. Scheper, and Guido C.H.E. De Croon. Vertical landing for micro air vehicles using event-based optical flow.Journal of Field Robotics, 35 (1):69–90, 2018. 2

    2018

  19. [25]

    Mc-veo: A visual-event odometry with accurate 6- dof motion compensation

    Jiafeng Huang, Shengjie Zhao, Tianjun Zhang, and Lin Zhang. Mc-veo: A visual-event odometry with accurate 6- dof motion compensation. IEEE Transactions on Intelligent Vehicles, 9(1):1756–1767, 2023. 2

    2023

  20. [26]

    Progres- sive spatio-temporal alignment for efficient event-based mo- tion estimation

    Xueyan Huang, Yueyi Zhang, and Zhiwei Xiong. Progres- sive spatio-temporal alignment for efficient event-based mo- tion estimation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 1537– 1546, 2023. 2

    2023

  21. [27]

    Determining shape and mo- tion from monocular camera: A direct approach using nor- mal flows

    Tak-Wai Hui and Ronald Chung. Determining shape and mo- tion from monocular camera: A direct approach using nor- mal flows. Pattern Recognition, 48(2):422–437, 2015. 1

    2015

  22. [28]

    Event- based semantic-aided motion segmentation

    Chenao Jiang, Julien Moreau, and Franck Davoine. Event- based semantic-aided motion segmentation. In International Conference on Computer Vision Theory and Applications (VISAPP 2024), 2024. 2

    2024

  23. [29]

    Bridges Jr

    Cecil C. Bridges Jr. Hierarchical cluster analysis. Psycho- logical Reports, 18(3):851–854, 1966. 6

    1966

  24. [30]

    Hanme Kim, Ankur Handa, Ryad Benosman, Sio-Hoi Ieng, and Andrew J. Davison. Simultaneous mosaicing and track- ing with an event camera. J. Solid State Circ , 43:566–576,

  25. [31]

    Hanme Kim, Stefan Leutenegger, and Andrew J. Davison. Real-time 3d reconstruction and 6-dof tracking with an event camera. In European Conference on Computer Vision, pages 349–364, 2016. 2

    2016

  26. [32]

    Spikeseg: Spiking segmentation via stdp saliency mapping

    Paul Kirkland, Gaetano Di Caterina, John Soraghan, and George Matich. Spikeseg: Spiking segmentation via stdp saliency mapping. In 2020 International Joint Conference on Neural Networks (IJCNN), pages 1–8, 2020. 2

    2020

  27. [33]

    Deep event visual odometry

    Simon Klenk, Marvin Motzet, Lukas Koestler, and Daniel Cremers. Deep event visual odometry. In 2024 International Conference on 3D Vision (3DV), pages 739–749, 2024. 2

    2024

  28. [34]

    Low-latency visual odometry using event- based feature tracks

    Beat Kueng, Elias Mueggler, Guillermo Gallego, and Da- vide Scaramuzza. Low-latency visual odometry using event- based feature tracks. In 2016 IEEE/RSJ International Con- ference on Intelligent Robots and Systems (IROS), pages 16– 23, 2016. 2

    2016

  29. [35]

    A 128 × 128 120 db 15µs latency asynchronous temporal con- trast vision sensor

    Patrick Lichtsteiner, Christoph Posch, and Tobi Delbruck. A 128 × 128 120 db 15µs latency asynchronous temporal con- trast vision sensor. IEEE Journal of Solid-State Circuits, 43 (2):566–576, 2008. 1

    2008

  30. [36]

    6-dof pose relocalization for event cameras with entropy frame and attention networks

    Hu Lin, Meng Li, Qianchen Xia, Yifeng Fei, Baocai Yin, and Xin Yang. 6-dof pose relocalization for event cameras with entropy frame and attention networks. In Proceedings of the 18th ACM SIGGRAPH International Conference on Virtual-Reality Continuum and its Applications in Ind...

    2022

  31. [37]

    Ground moving vehicle detection and movement tracking based on the neuromorphic vision sensor

    Xiangyong Liu, Guang Chen, Xuesong Sun, and Alois Knoll. Ground moving vehicle detection and movement tracking based on the neuromorphic vision sensor. IEEE In- ternet of Things Journal, 7(9):9026–9039, 2020. 2

    2020

  32. [38]

    Event-based visual inertial velometer

    Xiuyuan Lu, Yi Zhou, Junkai Niu, Sheng Zhong, and Shaojie Shen. Event-based visual inertial velometer. arXiv preprint arXiv:2311.18189, 2023. 2

    2023 arXiv

  33. [39]

    Continuous pose for monocular cameras in neural im- plicit representation

    Qi Ma, Danda Pani Paudel, Ajad Chhatkuli, and Luc Van Gool. Continuous pose for monocular cameras in neural im- plicit representation. In Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 5291–5301, 2024. 2

    2024

  34. [40]

    Data-driven feature tracking for event cameras

    Nico Messikommer, Carter Fang, Mathias Gehrig, and Da- vide Scaramuzza. Data-driven feature tracking for event cameras. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 5642– 5651, 2023. 2

    2023

  35. [41]

    Towards visual slam with event-based cam- eras

    Michael Milford, Hanme Kim, Stefan Leutenegger, and An- drew Davison. Towards visual slam with event-based cam- eras. In The problem of mobile sensors workshop in conjunc- tion with RSS, 2015. 2

    2015

  36. [42]

    Ev-imo: Motion seg- mentation dataset and learning pipeline for event cameras

    Anton Mitrokhin, Chengxi Ye, Cornelia Ferm ¨uller, Yian- nis Aloimonos, and Tobi Delbruck. Ev-imo: Motion seg- mentation dataset and learning pipeline for event cameras. In 2019 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 6105–6112, 2019. 2

    2019

  37. [43]

    Learning visual motion segmentation using event surfaces

    Anton Mitrokhin, Zhiyuan Hua, Cornelia Ferm ¨uller, and Yiannis Aloimonos. Learning visual motion segmentation using event surfaces. In Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 14414–14423, 2020. 2

    2020

  38. [44]

    Towards evasive maneuvers with quadrotors using dynamic vision sensors

    Elias Mueggler, Nathan Baumli, Flavio Fontana, and Davide Scaramuzza. Towards evasive maneuvers with quadrotors using dynamic vision sensors. In 2015 European Conference on Mobile Robots (ECMR), pages 1–8, 2015. 2

    2015

  39. [45]

    Lifetime estimation of events from dynamic vision sensors

    Elias Mueggler, Christian Forster, Nathan Baumli, Guillermo Gallego, and Davide Scaramuzza. Lifetime estimation of events from dynamic vision sensors. In 2015 IEEE International Conference on Robotics and Automation (ICRA), pages 4874–4881, 2015. 2

    2015

  40. [46]

    Dotie––detecting objects through temporal isolation of events using a spiking architecture

    Manish Nagaraj, Chamika Mihiranga Liyanagedera, and Kaushik Roy. Dotie––detecting objects through temporal isolation of events using a spiking architecture. In 2023 IEEE International Conference on Robotics and Automation (ICRA), pages 4858–4864, 2023. 2

    2023

  41. [47]

    Shahriar Negahdaripour and Berthold K.P. Horn. Direct pas- sive navigation. IEEE Transactions on Pattern Analysis and Machine Intelligence, (1):168–176, 1987. 1

    1987

  42. [48]

    Caldwell, and Nikos G

    Anh Nguyen, Thanh-Toan Do, Darwin G. Caldwell, and Nikos G. Tsagarakis. Real-time 6dof pose relocalization for event cameras with stacked spatial lstm networks. In Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, pages 0–0, 2019. 2

    2019

  43. [49]

    Parameshwara, Nitin J

    Chethan M. Parameshwara, Nitin J. Sanket, Chahat Deep Singh, Cornelia Ferm ¨uller, and Yiannis Aloimonos. 0- mms: Zero-shot multi-motion segmentation with a monocu- lar event camera. In 2021 IEEE International Conference on Robotics and Automation (ICRA), pages 9594–9600, 2021. 2

    2021

  44. [50]

    Toward an insect-inspired event-based au- topilot combining both visual and control events

    Thibaut Raharijaona, Julien Serres, Erik Vanhoutte, and Franck Ruffier. Toward an insect-inspired event-based au- topilot combining both visual and control events. In2017 3rd International Conference on Event-Based Control, Commu- nication and Signal Processing (EBCCSP), pages...

    2017

  45. [51]

    Evo: A geometric approach to event- based 6-dof parallel tracking and mapping in real time.IEEE Robotics and Automation Letters, 2(2):593–600, 2016

    Henri Rebecq, Timo Horstsch ¨afer, Guillermo Gallego, and Davide Scaramuzza. Evo: A geometric approach to event- based 6-dof parallel tracking and mapping in real time.IEEE Robotics and Automation Letters, 2(2):593–600, 2016. 2

    2016

  46. [52]

    Real-time visual-inertial odometry for event cameras using keyframe-based nonlinear optimization, 2017

    Henri Rebecq, Timo Horstschaefer, and Davide Scaramuzza. Real-time visual-inertial odometry for event cameras using keyframe-based nonlinear optimization, 2017. University of Zurich. 2

    2017

  47. [54]

    Sanket, Chethan M

    Nitin J. Sanket, Chethan M. Parameshwara, Chahat Deep Singh, Ashwin V . Kuruttukulam, Cornelia Ferm¨uller, Davide Scaramuzza, and Yiannis Aloimonos. Evdodgenet: Deep dynamic obstacle dodging with event cameras. In 2020 IEEE International Conference on Robotics and Automation (...

    2020

  48. [55]

    Event-based visual guidance inspired by honeybees in a 3d tapered tunnel

    Julien Serres, Thibaut Raharijaona, Erik Vanhoutte, and Franck Ruffier. Event-based visual guidance inspired by honeybees in a 3d tapered tunnel. In 2016 Second Interna- tional Conference on Event-Based Control, Communication, and Signal Processing (EBCCSP), pages 1–4, 2016. 2

    2016

  49. [56]

    Simultaneous opti- cal flow and segmentation (sofas) using dynamic vision sen- sor

    Timo Stoffregen and Lindsay Kleeman. Simultaneous opti- cal flow and segmentation (sofas) using dynamic vision sen- sor. arXiv preprint arXiv:1805.12326, 2018. 2

    2018 arXiv

  50. [58]

    Deep learning for pose estimation from event camera

    Ahmed Tabia, Fabien Bonardi, and Samia Bouchafa. Deep learning for pose estimation from event camera. In 2022 In- ternational Conference on Digital Image Computing: Tech- niques and Applications (DICTA), pages 1–7, 2022. 2

    2022

  51. [59]

    Event-based parti- cle filtering for robot self-localization

    David Weikersdorfer and J ¨org Conradt. Event-based parti- cle filtering for robot self-localization. In 2012 IEEE Inter- national Conference on Robotics and Biomimetics (ROBIO), pages 866–870, 2012. 2

    2012

  52. [60]

    Simultaneous localization and mapping for event-based vi- sion systems

    David Weikersdorfer, Raoul Hoffmann, and J ¨org Conradt. Simultaneous localization and mapping for event-based vi- sion systems. In International Conference on Computer Vi- sion Systems, pages 133–142, 2013. 2

    2013

  53. [61]

    A real-time event-based normal flow estimator

    Dehao Yuan and Cornelia Ferm ¨uller. A real-time event-based normal flow estimator. arXiv preprint arXiv:2504.19417, 2025. 2, 3

    2025 arXiv

  54. [62]

    Learn- ing normal flow directly from event neighborhoods

    Dehao Yuan, Levi Burner, Jiayi Wu, Minghui Liu, Jingxi Chen, Yiannis Aloimonos, and Cornelia Ferm ¨uller. Learn- ing normal flow directly from event neighborhoods. arXiv preprint arXiv:2412.11284, 2024. 1, 2, 3, 5

    2024 arXiv

  55. [63]

    A multi-scale re- current framework for motion segmentation with event cam- era

    Shaobo Zhang, Lei Sun, and Kaiwei Wang. A multi-scale re- current framework for motion segmentation with event cam- era. IEEE Access, 11:80105–80114, 2023. 2

    2023

  56. [64]

    Spike-based motion estimation for object tracking through bio-inspired unsupervised learning

    Yajing Zheng, Zhaofei Yu, Song Wang, and Tiejun Huang. Spike-based motion estimation for object tracking through bio-inspired unsupervised learning. IEEE Transactions on Image Processing, 32:335–349, 2022. 2

    2022

  57. [65]

    Event-based motion segmentation with spatio- temporal graph cuts

    Yi Zhou, Guillermo Gallego, Xiuyuan Lu, Siqi Liu, and Shaojie Shen. Event-based motion segmentation with spatio- temporal graph cuts. IEEE Transactions on Neural Networks and Learning Systems, 34(8):4868–4880, 2021. 2

    2021

  58. [66]

    Event-free moving object segmentation from moving ego vehicle

    Zhuyun Zhou, Zongwei Wu, Danda Pani Paudel, R´emi Bout- teau, Fan Yang, Luc Van Gool, Radu Timofte, and Do- minique Ginhac. Event-free moving object segmentation from moving ego vehicle. In 2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages ...

    2024

  59. [67]

    Unsupervised event-based learning of optical flow, depth, and egomotion

    Alex Zihao Zhu, Liangzhe Yuan, Kenneth Chaney, and Kostas Daniilidis. Unsupervised event-based learning of optical flow, depth, and egomotion. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 989–997, 2019. 2

    2019

  60. [68]

    Continuous-time object segmentation using high temporal resolution event camera

    Lin Zhu, Xianzhang Chen, Lizhi Wang, Xiao Wang, Yonghong Tian, and Hua Huang. Continuous-time object segmentation using high temporal resolution event camera. IEEE Transactions on Pattern Analysis and Machine Intelli- gence, 2024. 2

    2024

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.