Pith. sign in

REVIEW 4 major objections 4 minor 26 references

Locating Tennis Ball Impact on the Racket in Real Time Using an Event Camera

T0 review · 4 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read When it detects the ball's contour, an event-camera pipeline locates the tennis ball impact point on the racket to within 15 mm in under two seconds of computation.

desk verdict Genuinely new PATS indicator, honest about failures, but the abstract oversells and the 15 mm claim needs the missing calibration spelled out. read the letter →

arxiv 2506.08327 v1 pith:EZORFW4E submitted 2025-06-10 cs.CV

Yuto Kase , Kai Ishibe , Ryoma Yasuda , Yudai Washida , Sakiko Hashimoto This is my paper
classification cs.CV
keywords EventcameraTennisimpactlocationSportsanalyticsEvent-basedvisionsensorPolarityasymmetryintimesymmetryReal-timemeasurementRacketContourdetection
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 proposes a three-step pipeline that finds where a tennis ball strikes the racket from the raw event stream of an event camera, avoiding the memory burden of high-speed video and the labor of manual digitization. It claims that, when contour detection succeeds, the estimated impact position stays within 15 mm (less than a quarter of a ball diameter) of a visually inspected reference, inside the tolerance for analyzing tennis performance, and that each instance processes in under two seconds. The method first detects the swing interval from event-rate statistics, then finds the exact impact instant with an original event-based indicator called PATS, and finally fits ellipses to the ball and racket in event images at that instant. The paper also reports a sharp limit: under direct sunlight, string flicker adds events that make the ball contour undetectable in most instances.

What carries the argument

The machinery is PATS, the amount of polarity asymmetry in time symmetry. For each candidate impact time $t$, it splits the event packet into events before and after $t$, time-reverses the after-events, builds polarity event images and focal-time-weighted surfaces for both halves, and computes $\rho_t = \sum_{x,y} \lvert F^+_{\mathrm{prev}} * G_{\mathrm{prev}} \rvert \, \lvert F^-_{\mathrm{next}} * G_{\mathrm{next}} \rvert$. Because the ball is assumed brighter than the background, its bounce should show a $+$-to-$-$ flip across $t$, so $\rho_t$ peaks at the true impact instant. The focal time function uses Pattern 3, which weights recent events more but down-weights the very latest events, matching the roughly 4000 µs contact time while suppressing the under-100 µs string flicker.

What would settle it

Mount a racket on a fixture and launch a ball at a known grid of positions under uniformly lit, low-flicker conditions; if any successful contour run deviates by more than 15 mm from the known point, or if a bright-ball scene yields a timing error above 2000 µs, the central claim is wrong. A second check is to use a dark ball on a bright background and observe whether the polarity asymmetry inverts and timing detection fails, which would confirm that the method rests on the brightness-contrast assumption.

Watch

Extended reading notes

Core claim

The central discovery is that the instant of ball-racket impact leaves a characteristic asymmetric polarity signature in the event stream, and that this signature can be isolated by a quantity the paper calls PATS, the amount of polarity asymmetry in time symmetry. For a packet of events centered on a candidate time $t$, the method splits the packet into events before and after $t$, time-reverses the after-packet, builds polarity event images and focal-time-weighted surfaces for both halves, and sums the pixelwise product of the convolved plus-before and minus-after images. The impact time is the peak of this summed response, using the assumption that the ball is brighter than the background so its arrival flips pixels from $+$ to $-$. Once the impact time is known, binary event images at that instant are closed morphologically and fitted with ellipses to locate the ball relative to the racket. In 50 real rally instances the impact timing was found correctly in 46, and in every instance where contour detection succeeded the position error was below 12.1 percentage points in the racket's $u$ direction and 9.1 in $v$, which the paper translates to under 15 mm.

Load-bearing premise

The method depends on the ball being brighter than the background so that impact produces a clean bright-to-dark polarity flip at the pixel; direct sunlight can destroy that contrast through string flicker, and the paper's own numbers show the contour step failing in 17 of 20 sunny instances.

Editorial extensions

If this is right

  • A player can capture an entire practice session with a low-memory event camera and get per-shot impact locations, replacing high-speed video storage and manual digitization.
  • The measured distribution of impact positions across a player's shots can support personalized racket design, since the paper's data show the positions vary systematically among players.
  • Because processing finishes in under two seconds per instance, the system can deliver feedback between consecutive shots in a rally rather than after offline analysis.
  • PATS keeps impact-timing detection accurate even under direct sunlight, so the bottleneck in sunlight is specifically the contour step, not the detection of the impact instant.
  • The relevant accuracy guarantee is conditional on both ellipses being fitted; when contour detection fails, the overall pipeline produces no location output at all.

Reading between the lines

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

  • Inference: the same before/after polarity-asymmetry signature should appear in other ball-and-implement sports such as table tennis, badminton, or baseball whenever the ball contrasts with the background, so PATS may transfer to those settings with only threshold retuning.
  • Inference: using PATS as a hardware trigger to wake a frame-based or high-speed camera could combine the memory advantages of event sensors with the richer appearance of conventional images at the single most important instant of the swing.
  • Inference: the 15 mm accuracy figure is conditional on successful contour detection; counting the 17 sunny contour failures as errors yields an overall success rate near 27 of 50 instances, so a product-level metric should report contour failure as part of the error budget.
  • Inference: the swing-interval thresholds are fixed from only ten training shots; an adaptive or per-player thresholding scheme is a testable extension that could remove the two swing-range overestimations and the three timing misestimations reported on the test set.
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 / 4 minor

Summary. The paper proposes a three-step pipeline to locate the tennis ball impact position on the racket from event-camera data: (1) detection of the swing time range via event-rate thresholds, (2) detection of the impact instant via a new indicator called PATS (polarity asymmetry in time symmetry), and (3) contour extraction of the ball and racket via binary event images and ellipse fitting. The output is the ball position relative to the racket axes expressed as percentages. Experiments were carried out on 50 test instances from 5 players, with players 4 and 5 under direct sunlight. The paper reports that for all instances where contour detection succeeded (27 of 46 instances that reached that step), the relative absolute difference was below 12.1 %pt in u and 9.1 %pt in v, which it equates to less than 15 mm, and that mean computation time per instance was under 2 seconds.

Significance. If the accuracy claim holds, the method provides a memory-efficient, high-speed approach to measuring racket impact locations, which is relevant for personalized equipment design and player analysis. The PATS indicator is a genuinely original event-based signal for impact timing and appears to work under the tested conditions, including some sunlight cases. The main weaknesses are that the paper overstates the overall success in the abstract (contour detection failed in 17 of 20 direct-sunlight instances), the 15 mm equivalence is not justified by a reported pixel-to-millimeter calibration, and the visual ground truth is not independently validated. These issues are addressable with additional reporting rather than a fundamental redesign of the method.

major comments (4)
  1. [Section 5.1, Table 2] The claim that relative absolute differences below 12.1 %pt (u) and 9.1 %pt (v) correspond to less than 15 mm is not supported by the paper. The output is expressed as percentages of the racket ellipse axes (Section 3.5), but neither the physical dimensions of the racket head nor the camera calibration mapping pixels to millimeters is reported. With a typical racket head length of about 320 mm, a 12.1 %pt error could correspond to tens of millimeters, so the stated equivalence is unverifiable unless a specific scale factor is disclosed. The paper must report the actual racket dimensions and the conversion used.
  2. [Abstract and Section 5.2] The abstract states that "The results of the experiments were within the permissible range for measuring tennis players' performance" without qualification, but under direct sunlight contour detection succeeded in only 3 of 20 instances (Table 2, players 4 and 5). The paper's own limitation section acknowledges this failure. The abstract and conclusion should explicitly restrict the accuracy claim to instances where contour detection succeeded, e.g., "for all instances where contour detection was successful."
  3. [Table 2] The "Time Range of Swing" rows for players No. 4 and 5 are listed as 11/10, which is arithmetically impossible because each player contributed 10 test instances. The text of Section 4.5 says one instance each of overestimating the swing range occurred, so these entries need correction and the counting convention (overestimate counted as success or failure) should be clarified across the table.
  4. [Section 4.3] The ground truth for both impact timing and position is defined by the authors' own visual inspection, with no inter-rater reliability measure or independent validation. The "permissible range" of 15 mm is also taken from the authors' prior work [26]. The paper should either provide evidence that the visual ground truth is reproducible (e.g., multiple annotators) or explicitly discuss the self-referential nature of the accuracy metric as a limitation.
minor comments (4)
  1. [Section 3.5] The definition of the u/v coordinate system is ambiguous: it is not stated whether the reported percentages are relative to the semi-axis lengths or the full axis lengths of the racket ellipse, which is essential for interpreting the accuracy results.
  2. [Section 4.5 / Table 2] The notation "%pt" is used inconsistently and is not formally defined in the metrics section; the paper should define whether the relative absolute difference is an absolute difference in percentage points or a relative ratio.
  3. [Section 1 and 5.1] The "real-time" claim is based on offline processing of pre-recorded event streams with a mean computation time below 2 seconds per instance; the paper does not demonstrate streaming or online processing, so the terminology should be softened to "near-real-time" or "fast enough for post-hoc analysis of rallies."
  4. [Figure 4 caption] The sentence "Image (c - ii) detected more ball events than image (c - i)" is unclear; a caption should describe what is being compared rather than using the word "detected," since the images are visualizations, not detector outputs.

Circularity Check

1 steps flagged · score 2.0 of 10

The method is internally non-circular; the only self-referential element is the evaluation criterion borrowed from the authors' own prior work (Yasuda et al. [26]), which does not force the measured accuracy or timing.

  1. other [Section 4.3 (Metrics) and Section 5.1 (Consideration), citing Reference [26]]
    "Section 4.3: "The visual definitions of the time and position at impact using an event camera are based on those of Yasuda et al. [26]." Section 5.1: "For all instances where contour detection was successful, the relative absolute difference in the impact position was within the permissible range for measuring tennis players' performance as indicated by Yasuda et al. [26]. This range corresponds to less than one-quarter of the diameter of a tennis ball (15 mm).""

    The threshold used to declare success (the 15 mm "permissible range") and the visual ground-truth definitions of impact timing/position are both taken from reference [26], whose author list includes Yasuda, Kase, Ishibe, Washida, and Hashimoto, i.e., the present authors. Thus the headline claim "within the permissible range" is evaluated against a standard set by the same research group rather than an independently established benchmark. This is a mild self-referential evaluation setup. It is not a derivation-level circularity, however: the reported %pt differences and computation times are actual measurements on held-out test data, and the algorithm's outputs do not set the threshold.

full rationale

The derivation chain is self-contained. PATS in Eq. (11) is computed from polarity-symmetric event images and time surfaces; it is not fitted to the evaluation labels, and the impact timing is detected from the peak of the ρt series before contours are estimated. The contour and ellipse steps use standard OpenCV operations, and the output is a normalized coordinate on the racket ellipse. The train/test split (first two balls per player for tuning the variables in Table 1; the remaining ten balls for evaluation) means the variables are not fitted to the test instances. The only self-referential element is that the ground-truth visual definitions and the "permissible range" of 15 mm come from reference [26], a prior paper by the same group. That makes the evaluation criterion group-defined, but it does not make the measured %pt errors or computation times equivalent to an input by construction. Separately, the conversion from %pt to millimeters is not documented, which is a reproducibility and correctness concern rather than circularity. Accordingly, the circularity score is 2 rather than higher.

Assumptions & free parameters 12 free parameters · 5 assumptions · 0 invented entities

The central claim rests on a set of thresholds and accumulation times fitted to 10 training instances, plus domain assumptions about racket motion and ball brightness. No code, data, or sensitivity analysis is provided.

free parameters (12)
  • tacc (swing) = 500
    Accumulation time for event packets in swing detection, set using training data (Table 1).
  • tstrd (swing) = 500
    Stride time for event rate sampling, set using training data (Table 1).
  • n_epsilon = 10
    Number of consecutive event rates for mean/variance computation, set using training data (Table 1).
  • tau_mean = 1e7
    Mean threshold for swing start, set using training data (Table 1).
  • tau_var = 6e11
    Variance threshold for swing start, set using training data (Table 1).
  • tau_t = 100000
    Minimum time after start before swing end can be declared, set using training data (Table 1).
  • tacc (impact) = 4000
    Accumulation time for PATS event packets, set using training data (Table 1).
  • tstrd (impact) = 500
    Stride time for PATS sampling, set using training data (Table 1).
  • nc = 3
    Number of candidate impact peaks considered for centroid selection, set using training data (Table 1).
  • tacc (ball) = 2000
    Accumulation time for ball contour event image, set using training data (Table 1).
  • tacc (racket) = 500
    Accumulation time for racket contour event image, set using training data (Table 1).
  • focal time function pattern = Pattern 3
    Chosen from Fig. 6 to suppress string flicker; selected using training data (Section 4.4).
assumptions (5)
  • domain assumption The ball is brighter than the background, so impact creates a + then - polarity pattern.
    Section 3.3: 'We assume that the ball is brighter than the background.'
  • domain assumption The racket moves faster than other objects, so event rate rises during a swing.
    Section 3.2 relies on this to detect the swing time range.
  • domain assumption Tennis ball contact duration is about 4000 microseconds and string flicker about 100 microseconds.
    Section 3.3, citing Cross [2], used to design the focal time function.
  • domain assumption Visual inspection using Yasuda et al.'s definition of impact timing is a correct ground truth.
    Section 4.3 defines ground truth from the authors' own prior paper [26].
  • standard math OpenCV morphological closing and ellipse fitting can recover ball and racket contours from binary event images.
    Section 3.4 uses standard OpenCV operations without further derivation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Locating Tennis Ball Impact on the Racket in Real Time Using an Event Camera." pith.science (2026). https://pith.science/paper/EZORFW4E

@misc{pith2026250608327,
  author       = {Pith},
  title        = {Pith review of: Locating Tennis Ball Impact on the Racket in Real Time Using an Event Camera},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EZORFW4E}},
  note         = {Machine review of arXiv:2506.08327}
}
read the original abstract

In racket sports, such as tennis, locating the ball's position at impact is important in clarifying player and equipment characteristics, thereby aiding in personalized equipment design. High-speed cameras are used to measure the impact location; however, their excessive memory consumption limits prolonged scene capture, and manual digitization for position detection is time-consuming and prone to human error. These limitations make it difficult to effectively capture the entire playing scene, hindering the ability to analyze the player's performance. We propose a method for locating the tennis ball impact on the racket in real time using an event camera. Event cameras efficiently measure brightness changes (called `events') with microsecond accuracy under high-speed motion while using lower memory consumption. These cameras enable users to continuously monitor their performance over extended periods. Our method consists of three identification steps: time range of swing, timing at impact, and contours of ball and racket. Conventional computer vision techniques are utilized along with an original event-based processing to detect the timing at impact (PATS: the amount of polarity asymmetry in time symmetry). The results of the experiments were within the permissible range for measuring tennis players' performance. Moreover, the computation time was sufficiently short for real-time applications.

Figures

Figures reproduced from arXiv: 2506.08327 by the authors.

Figure 1
Figure 1. Overview of the proposed method. A set of events captured by an event camera while playing tennis are input; + polarity events are represented in blue and − polarity events are represented in red. Locating the ball impact on the racket involves three steps. The output is the visualization of the ball position, which is relative to the racket. To measure the impact location, conventional frame-based cameras (high￾spe… view at source ↗
Figure 2
Figure 2. Time series of event packets εt in tennis scenes. (a) event data ek colored according to polarity (+ in blue, − in red); the areas with smaller x-coordinates show clusters of events caused by racket swings, whereas the areas with larger x-coordinates show clusters of events caused by ball bounces. (b) time series of event rates (|εt| per second) which increase while a player is swinging the racket; this player swing… view at source ↗
Figure 3
Figure 3. Time series of the mean and variance for the consecutive nε event rates. (a) time series of the mean, where the red solid line represents the threshold τmean. (b) time series of the variance, where the solid red line represents the threshold τvar. The black dashed lines represent tstart and tend, respectively. τmean and τvar are used to identify tstart, while τmean is used for tend. 3.3 Timing at Impact As shown in … view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: Comparison of the visualization of PATS images before and at impact. (a) 3D plot of events ek near the impact timing timp, where the blue points represent +; the red points represent −; the black plane represents timp; and the orange edge represents a plane vertical to…
Figure 5
Figure 5. Figure 5: Timing at impact procedure. (a) 3D plot of events ek, where blue points rep￾resent +; red points represent −; and the black plane represents reference time t. Plot (a) represents εt; plot (a - i) represents εprev; plot (a - ii) represents εnext; and plot (a - ii’) repr…
Figure 6
Figure 6. Figure 6: Finally, the peak in the time series of ρt values is detected to identify timp, as shown in [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 6
Figure 6. Figure 6: Patterns of various focal time functions. Pattern 1 shows a function with uni￾form ratio (1.0), which is the same as an event image. Pattern 2 shows a function where the ratio increases linearly, which is equivalent to a time surface. Pattern 3 shows a function where t…
Figure 7
Figure 7. Figure 7: Peaks in the time series of ρt values (i.e. PATS). Graph (a) time series of ρt. Graph (a’) shows the Laplacian filtered time series, where the black solid line represents the estimated impact timing timp within the estimated time range of the swing (black dashed line).…
Figure 8
Figure 8. Figure 8: Ball and racket contour procedure. (a) binary images at the impact timing timp with different accumulation times tacc. Image (a - i) detects the racket with tacc = 500, and image (a - ii) detects the ball with tacc = 2000, where the red ellipse represents the detected …
Figure 9
Figure 9. Figure 9: Play scene and equipment for our experiments. The event camera captured the scene from behind the player, and the ball was tossed from the machine located in the front right, as shown in the left figure. In the case of left-handed players, the position of the machine w…
Figure 10
Figure 10. Figure 10: Failure of contour estimation due to string flickering under direct sunlight. (a) event image at the impact timing. (b) binary image converted from image (a). The flickering of the strings makes it difficult to identify the contour of the ball. however, the detection …

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

26 extracted references · 26 canonical work pages

  1. [26]

    In: Sports Informatics (SI)

    Yasuda, R., Kase, Y., Ishibe, K., Washida, Y., Hashimoto, S.: [a proposal for tennis impact location measurement method with an event camera] event camera wo mochiita koushiki tennis no daten ichi sokutei shuhou no teian (in Japanese). In: Sports Informatics (SI). vol. 2024, pp. 1–6 (2024)

    work page 2024

  2. [1]

    com/en/silkyevcam-hd/ Accessed on 2025-02-04

    CenturyArks: SilkyEvCam HD - CenturyArks Co., Ltd.,https://centuryarks. com/en/silkyevcam-hd/ Accessed on 2025-02-04

    work page 2025

  3. [2]

    Sports Engineering2(1), 23–33 (1999)

    Cross: Dynamic properties of tennis balls. Sports Engineering2(1), 23–33 (1999)

    work page 1999

  4. [3]

    In: 2024 IEEE International Workshop on Sport, Technology and Research (STAR)

    Deckyvere, A., Cioppa, A., Giancola, S., Ghanem, B., Droogenbroeck, M.V.: In- vestigating event-based cameras for video frame interpolation in sports. In: 2024 IEEE International Workshop on Sport, Technology and Research (STAR). pp. 138–143 (2024)

    work page 2024

  5. [4]

    IEEE Transactions on Pattern Analysis and Machine Intel- ligence 44(1), 154–180 (2022)

    Gallego, G., Delbrück, T., Orchard, G., Bartolozzi, C., Taba, B., Censi, A., Leutenegger, S., Davison, A.J., Conradt, J., Daniilidis, K., Scaramuzza, D.: Event- based vision: a survey. IEEE Transactions on Pattern Analysis and Machine Intel- ligence 44(1), 154–180 (2022)

    work page 2022

  6. [5]

    In: IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    Gallego, G., Rebecq, H., Scaramuzza, D.: A unifying contrast maximization frame- work for event cameras, with applications to motion, depth, and optical flow esti- mation. In: IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 3867–3876 (2018)

    work page 2018

  7. [6]

    In: IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops

    Gossard, T., Krismer, J., Ziegler, A., Tebbe, J., Zell, A.: Table tennis ball spin estimation with an event camera. In: IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops. pp. 3347–3356 (2024)

    work page 2024

  8. [7]

    In: 13th Conference of the International Sports Engineering Association (2020)

    Ikenaga, M., Okuma, N., Nishiyama, H., Chiba, S., Nishino, K., Omori, G., Nunome, H.: Influence of ball impact location on racquet kinematics, forearm mus- cle activation and shot accuracy during the forehand groundstrokes in tennis. In: 13th Conference of the International Sports Engineering Association (2020)

    work page 2020

Show all 26 references
  1. [8]

    com Accessed on 2025-02-04

    iniVation: iniVation − Neuromorphic vision systems (2023),https://inivation. com Accessed on 2025-02-04

    2023

  2. [9]

    Karditsas, H.E.: Large-Scale Method for Identifying the Relationships between RacketPropertiesandPlayingCharacteristics.SheffieldHallamUniversity(United Kingdom) (2020)

    2020

  3. [10]

    Kiraly, C., Merloti, P.: Golf club head measurement system, United States Patent US8951138B2, 2015-02-10

    2015

  4. [11]

    Kiraly, C., Wintriss, V.: Flight parameter measurement system, European Patent EP1509781B1, 2015-04-29

    2015

  5. [12]

    IEEE Transactions on Pattern Analysis and Machine Intelligence2017(7), 1346–1359 (2017)

    Lagorce, X., Orchard, G., Galluppi, F., Shi, B.E., Benosman, R.B.: Hots: A hier- archy of event-based time-surfaces for pattern recognition. IEEE Transactions on Pattern Analysis and Machine Intelligence2017(7), 1346–1359 (2017)

    2017

  6. [13]

    In: IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops

    Nakabayashi, T., Higa, K., Yamaguchi, M., Fujiwara, R., Saito, H.: Event-based ball spin estimation in sports. In: IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops. pp. 3367–3375 (2024)

    2024

  7. [14]

    In: 6th International Workshop on Multimedia Content Analysis in Sports

    Nakabayashi, T., Kondo, A., Higa, K., Girbau, A., Satoh, S., Saito, H.: Event- based high-speed ball detection in sports video. In: 6th International Workshop on Multimedia Content Analysis in Sports. pp. 55–62 (2023)

    2023

  8. [15]

    In: 16 International Symposium on Biomechanics in Sports (1998)

    Naß, D., Hennig, E.M., Schnabel, G.: Ball impact location on a tennis racket head and its influence on ball speed, arm shock and vibration. In: 16 International Symposium on Biomechanics in Sports (1998)

    1998

  9. [16]

    opencv.org/4.x/d4/d86/group__imgproc__filter.html Accessed on 2025-02-04

    OpenCV: OpenCV Image Filtering #morphologyEx (2025), https://docs. opencv.org/4.x/d4/d86/group__imgproc__filter.html Accessed on 2025-02-04

    2025

  10. [17]

    OpenCV: OpenCV Structural Analysis and Shape Descriptors #fitEllipse (2025), https://docs.opencv.org/4.x/d3/dc0/group__imgproc__shape.html Accessed on 2025-02-04 Locating Tennis Ball Impact Using an Event Camera 17

    2025

  11. [18]

    ai Accessed on 2025-02-04

    Prophesee: Prophesee | Metavision Technologies (2025),https://www.prophesee. ai Accessed on 2025-02-04

    2025

  12. [19]

    Prophesee: SDK Core Algorithms − Metavision SDK Docs 5.1.0 documentation #BaseFrameGenerationAlgorithm (2025), https://docs.prophesee.ai/stable/ api/python/core/bindings.html Accessed on 2025-02-04

    2025

  13. [20]

    Prophesee: SDK CV Python bindings API — Metavision SDK Docs 5.1.0 docu- mentation #ActivityNoiseFilterAlgorithm (2025), https://docs.prophesee.ai/ stable/api/python/cv/bindings.html Accessed on 2025-02-04

    2025

  14. [21]

    In: European Conference on Computer Vision (ECCV)

    Shiba, S., Aoki, Y., Gallego, G.: Secrets of event-based optical flow. In: European Conference on Computer Vision (ECCV). pp. 628–645 (2022)

    2022

  15. [22]

    In: IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2018)

    Sironi, A., Brambilla, M., Bourdis, N., Lagorce, X., Benosman, R.: Hats: His- tograms of averaged time surfaces for robust event-based object classification. In: IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2018)

    2018

  16. [23]

    In: IEEE/CVF International Conference on Computer Vision (ICCV)

    Stoffregen, T., Gallego, G., Drummond, T., Kleeman, L., Scaramuzza, D.: Event- based motion segmentation by motion compensation. In: IEEE/CVF International Conference on Computer Vision (ICCV). pp. 7244–7253 (2019)

    2019

  17. [24]

    Tuxen, F.: System and method for determining impact characteristics of sports ball striking element, United States Patent US10953303B2, 2021-03-23

    2021

  18. [25]

    Yamashita, K., Matsunaga, H.: Sensor device, analyzing device, and recording medium for detecting the position at which an object touches another object, United States Patent US9551572B2, 2017-01-24

    2017

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.