Pith. sign in

REVIEW 5 major objections 5 minor 43 references

Developing Normative Gait Cycle Parameters for Clinical Analysis Using Human Pose Estimation

T0 review · 5 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read This paper derives normative joint-angle parameters for the human gait cycle from 351 RGB-video cycles of typical gait, and uses them to flag kinematic abnormalities in single and multiple joints.

desk verdict A clean pipeline for video-based joint-angle norms, but the clinical-norm claim outruns the evidence; worth review with major revision. read the letter →

arxiv 2411.13716 v1 pith:URUWIBFY submitted 2024-11-20 cs.CV

Rahm Ranjan , David Ahmedt-Aristizabal , Mohammad Ali Armin , Juno Kim This is my paper · ORCID
classification cs.CV
keywords gaitanalysishumanposeestimationnormativeparametersjointanglesclinicalRGBvideokinematicdeviationcycle
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

This paper attempts to show that a usable clinical norm for human gait can be built purely from monocular RGB video, without markers or wearable sensors. The authors pool 351 typical gait cycles from three video datasets, extract 2D joint angles for ten joints with a lightweight pose estimator, and normalize every cycle from heel strike to heel strike. They then define the normative parameter for each joint as the mean and standard deviation of these angles across the pooled cycles. On four unseen videos, joint angles that fall outside one standard deviation from the mean are flagged as abnormal, for single joints and multiple joints simultaneously. The value of the claim is that a clinician could screen gait deviations with an ordinary camera and get a temporally localised, explainable readout.

What carries the argument

The central object is the per-joint normative curve: for each of ten joints, a mean angle and a one-standard-deviation band across the 0–100% gait cycle, computed from 351 temporally normalised cycles. The mechanism that carries the argument is the vector-based joint angle calculation using the arctan2 function on the 2D keypoint coordinates, followed by cubic spline interpolation to align every cycle to a common time base. The normative mean and standard deviation serve as the reference distribution; a joint angle in a test video is deemed potentially abnormal when it lies outside one standard deviation of the mean at the corresponding cycle percentage. This statistical definition is what turns raw pose estimates into clinically interpretable flags.

What would settle it

Concrete test: record a separate group of healthy adults from the same left-side view, run the same pose-estimation and joint-angle pipeline, and count how many of their cycles fall outside the one-standard-deviation band. If a large fraction of known-normal cycles are flagged as abnormal, the normative band is too tight or the pooling assumption is wrong. A second decisive check is to compare the 2D normative curves against marker-based 3D motion capture on the same subjects; if the 2D-to-3D discrepancy is comparable to the width of the normal band, the band reflects camera geometry rather than biological variation.

Watch

Extended reading notes

Core claim

The paper's central claim is that a clinically representative normative parameter for each of ten joints can be derived from a large set of typical gait cycles recorded in RGB video, and that this parameter supports cycle-wise kinematic deviation and abnormality detection. The normative curves are built by computing 2D joint angles from BlazePose keypoints, mapping each frame to a percentage of the gait cycle bounded by the left heel strike, and interpolating missing frames with cubic splines. The publication states that these mean and standard deviation curves give clinicians reference values to compare a patient's joint angles against a normative population using only monocular video. Demonstrations on four test videos show that atypical gait produces more detected abnormalities than typical gait, and that the degree of deviation can be visualised as shading over the cycle.

Load-bearing premise

The load-bearing premise is that 351 gait cycles pooled from three video datasets by different cameras and subjects, with multiple cycles from the same subject, can be treated as one exchangeable population of normal gait, so the pooled mean and standard deviation curves represent a clinically meaningful norm.

Editorial extensions

If this is right

  • A clinician could screen a patient's gait from a single lateral video and immediately see which joints deviate from the norm and at which phase of the gait cycle.
  • Because the analysis is cycle-wise, the same pipeline can track a patient over time, comparing each new cycle against the same normative band to monitor change during rehabilitation.
  • Joint angles are a universal kinematic measure, so the normative-curve approach is not locked to gait; the same machinery could be applied to other actions once cycles are defined.
  • The severity visualisation gives a prioritisation cue: darker regions indicate larger deviations, which can help clinicians focus attention on the most affected joints.
  • This establishes a baseline normative reference where none existed for RGB-video-only gait analysis, which is a prerequisite for automated abnormality screening.

Reading between the lines

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

  • If the pooling assumption holds, the same pipeline could build norms for other actions, since joint angles are action-agnostic and the cycle definition could be adapted.
  • A validation study against clinician ratings on a larger set of videos would likely tune the one-standard-deviation threshold, since 'abnormal' in clinical practice is not a fixed statistical distance.
  • Because multiple cycles come from the same subjects, the reported standard deviations probably understate between-subject variability; a random-effects model treating subject as a cluster would give a more honest norm.
  • The left-side-view-only construction means the norms are viewpoint-specific; a direct test would be to compute norms from one dataset and evaluate abnormality detection on the others without pooling.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 5 minor

Summary. The paper proposes a data-driven pipeline for building normative gait-cycle joint-angle parameters from monocular RGB video, using BlazePose keypoints and a 2D vector-based angle computation (Eqs. 1–2). The authors pool 351 'normal' gait cycles from three datasets (GPJATK, GAVD, CASD), temporally normalize them to percent gait cycle, interpolate with cubic splines, and compute per-joint mean and standard-deviation bands as normative values. They then flag any angle outside one standard deviation as potentially abnormal, visualize single- and multi-joint deviations, and demonstrate the approach on four videos, claiming cycle-wise kinematic deviation and abnormality detection for clinical use.

Significance. If the normative parameters were properly validated, the approach could provide a low-cost, accessible alternative to marker-based gait analysis, with the practical advantage of using only a single RGB camera. The use of joint angles—an established clinical kinematic measure—and the emphasis on cycle-wise, phase-localized abnormality detection are commendable. The paper also ships a clear pipeline description that is easy to reproduce from the text. However, the clinical significance claimed by the authors is presently unsupported: the normative values are unvalidated sample statistics from a small, heterogeneous, and uncalibrated dataset, and the evaluation is qualitative and circular. The contribution is best read as a baseline framework, not as clinically usable normative values.

major comments (5)
  1. [Section IV.E / Table I / Section V-C] The central claim of 'clinically representative normative values' rests on pooling uncalibrated 2D joint angles from videos captured with different camera geometries across three datasets. With a single lateral view and no camera calibration (acknowledged in Section V-C), the 2D angle from Eq. (1) depends on camera height, pan, and distance, not only on joint kinematics. If camera geometry differs across datasets, the pooled mean and standard deviation curves are a mixture of dataset-specific offsets. The paper provides no per-dataset mean curves, no between-dataset variance comparison, and no validation against marker-based motion capture. Without evidence that these 2D angles are comparable across datasets, the normative values do not support the clinical claim.
  2. [Section V.A / Table III / Section IV.E] The four test videos are drawn from the same three datasets that define the normative bands (Table I vs. Table III), which makes the evaluation circular: a 'typical' test video largely falls near a mean it helped create. More seriously, test video four uses subject 3 from CASD, and the same subject's normal gait cycles contribute to the normative pool, so the same individual is used to define normality and then evaluated as atypical. The reported detection results therefore conflate within-subject and between-subject variability and cannot demonstrate abnormality detection against a normative population. The authors should evaluate on held-out subjects or an external dataset.
  3. [Section V.B / Fig. 2] The abnormality detection threshold is set to one standard deviation from the mean with no clinical justification or sensitivity analysis. Under Gaussian assumptions, this threshold flags roughly 32% of a normal population's cycles as 'abnormal' by construction, so the observed red dots are not evidence of gait abnormality. The paper does not show that one standard deviation corresponds to a clinically meaningful threshold, nor does it test how the detection output changes with alternative thresholds. Without such analysis, the claimed ability to 'detect potential deviations' is not established.
  4. [Section V.B / Section V-C] The experimental evaluation is entirely qualitative: four videos, visual inspection of red/blue dots, and no quantitative accuracy metrics, no statistical comparison between typical and atypical conditions, and no comparison to a reference standard. The claim that 'atypical gait patterns' produce 'an observable increase' in deviations (Section V.B.2) is supported only by eyeballing Figures 2–6. The paper needs quantitative evaluation, e.g., sensitivity/specificity for classifying held-out typical vs. atypical cycles, or a per-cycle deviation score with a statistical test.
  5. [Section IV.A / Table I] The sample counts are internally inconsistent. The text states GPJATK has 42 subjects/99 cycles and GAVD has 11 subjects/71 cycles, while Table I swaps these values (GPJATK 11/71, GAVD 42/99). In addition, the text reports 137 complete normal gait cycles for CASD, but Table I lists 181. These discrepancies are load-bearing because the normative curves and any pooled statistics depend on the exact composition of the dataset; they must be resolved and the counts verified against the original dataset documentation.
minor comments (5)
  1. [Section IV.E] The authors describe the normative pool as 'a large RGB video dataset of typical gait cycles,' but the dataset comprises 73 subjects and 351 cycles, many from the same subjects and captured in controlled conditions. 'Large' is an overstatement for a clinical-norm claim; consider a more modest description such as 'a multi-dataset pool of 351 normal cycles.'
  2. [Section IV.E] The method assigns each frame a position within the gait cycle, but the text does not describe how the cycle percentage is computed from the annotated heel-strike events. It would clarify the pipeline to state whether linear interpolation between annotated start and end frames or a more sophisticated phase estimation is used.
  3. [Fig. 12 caption] The caption says 'outside of standard deviation of mean' but should read 'outside one standard deviation of the mean' for consistency with Fig. 11 and the main text.
  4. [Throughout] The dataset name is inconsistently written as 'GAVD' in most places but 'GA VD' in Sections IV.A.2 and Fig. 8; unify the spelling.
  5. [Section V-C] The limitation paragraph describes the dataset as 'uncontrolled in nature,' but GPJATK and CASD are controlled indoor recordings per Section IV.A. This contradicts the earlier description and should be reworded to refer only to the GAVD component or to the overall heterogeneity.

Circularity Check

1 steps flagged · score 6.0 of 10

Normative bands are built from the same datasets and, for test videos 3 and 4, the same CASD subject used for 'unseen' test videos, so the typical/atypical detection reduces to an in-sample consistency check rather than an independent prediction.

  1. fitted input called prediction [Section IV.E and Section V.A]
    "A clinically representative normative parameter for each joint is derived from a large RGB video dataset of typical gait cycles... statistical measures—mean and standard deviation—are calculated from the interpolated data, providing normative values and tolerances to determine typical joint angle measures. ... To demonstrate cycle-wise kinematic deviation and abnormality detection using our parameters, we test on four randomly selected unseen gait videos sourced from the various aforementioned datasets (See Table I). ..."

    The normative mean/SD curves are computed by pooling all 351 normal cycles from GPJATK, GAVD and CASD (Section IV.A/IV.E). The four 'test' videos are drawn from those same three datasets, and test videos 3 and 4 use the same CASD subject whose normal gait cycles are part of the 137 CASD normal cycles contributing to the norm. Consequently, the typical test video is compared against a norm that includes its own subject's data, so a near-mean curve is expected by construction rather than predicted. The atypical video is contrasted with that same subject's own normal cycles, making the abnormality flag a within-subject instructed contrast rather than evidence of a generalizable normative detector.

full rationale

The construction of the normative curves themselves is not circular: a mean and standard deviation computed from pooled joint-angle curves is a descriptive summary of the input data, not a hidden restatement of the outputs. The circularity enters at the validation step. The paper calls four videos 'unseen' and uses them to demonstrate abnormality detection, but those videos are sourced from the same datasets that define the normative bands, and in the clearest case (CASD subject 3) the same individual's normal gait is inside the norm against which their typical and atypical gaits are judged. This makes the reported typical/atypical separation an in-sample consistency check. I also note, as a correctness/transparency issue rather than circularity, that Section IV.A text reports GPJATK as 42 subjects/99 cycles and GAVD as 11 subjects/71 cycles, while Table I swaps these values and lists CASD cycles as 181 versus 137 in the text; this undermines confidence in the pooling description but does not itself constitute a circular derivation.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The central contribution depends on several domain assumptions about pose-estimation accuracy and comparability of uncalibrated 2D angles, plus an arbitrary threshold. These assumptions are flagged in the paper only as limitations, not tested.

free parameters (2)
  • Abnormality threshold = 1 standard deviation
    Chosen in Section V-B.1 without clinical justification; it directly determines which data points are labeled abnormal and drives all reported detection results.
  • Gait cycle phase resolution = 100 percent bins (0% to 100%)
    The cycle is re-sampled onto a fixed percent timeline via cubic spline interpolation; the choice of binning and interpolation settings affects the shape of the normative curves, though it is a standard normalization.
assumptions (4)
  • domain assumption BlazePose keypoint detections are sufficiently accurate for clinical joint-angle measurement in these videos.
    Invoked in Section IV.C; relies on prior validation studies [38], [39] rather than a within-paper accuracy check.
  • domain assumption A single lateral camera view with no calibration yields joint angles comparable across subjects and datasets.
    Stated as a limitation in Section V-C; the normative pooling treats all 2D angles as directly comparable, which the paper admits is compromised by perspective and parallax.
  • ad hoc to paper Pooled gait cycles are treated as independent and identically distributed samples of a single normal-gait population.
    Section IV.E averages over all cycles without modeling subject-level or dataset-level clustering, so the reported standard deviations mix inter-subject, inter-dataset, and intra-subject variation.
  • ad hoc to paper Mean plus or minus one standard deviation defines the clinically relevant normal range.
    Adopted in Section V-B.1 with no calibration against clinical outcomes or established norms.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Developing Normative Gait Cycle Parameters for Clinical Analysis Using Human Pose Estimation." pith.science (2026). https://pith.science/paper/URUWIBFY

@misc{pith2026241113716,
  author       = {Pith},
  title        = {Pith review of: Developing Normative Gait Cycle Parameters for Clinical Analysis Using Human Pose Estimation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/URUWIBFY}},
  note         = {Machine review of arXiv:2411.13716}
}
read the original abstract

Gait analysis using computer vision is an emerging field in AI, offering clinicians an objective, multi-feature approach to analyse complex movements. Despite its promise, current applications using RGB video data alone are limited in measuring clinically relevant spatial and temporal kinematics and establishing normative parameters essential for identifying movement abnormalities within a gait cycle. This paper presents a data-driven method using RGB video data and 2D human pose estimation for developing normative kinematic gait parameters. By analysing joint angles, an established kinematic measure in biomechanics and clinical practice, we aim to enhance gait analysis capabilities and improve explainability. Our cycle-wise kinematic analysis enables clinicians to simultaneously measure and compare multiple joint angles, assessing individuals against a normative population using just monocular RGB video. This approach expands clinical capacity, supports objective decision-making, and automates the identification of specific spatial and temporal deviations and abnormalities within the gait cycle.

Figures

Figures reproduced from arXiv: 2411.13716 by the authors.

Figure 1
Figure 1. Example of typical kinematic parameters of left knee [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 4
Figure 4. Multi-joint abnormality detection for test video [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figure 5
Figure 5. Multi-joint abnormality detection for test video [PITH_FULL_IMAGE:figures/full_fig_p005_5.png] view at source ↗
Figures from the paper (7 more)
Figure 6
Figure 6. Figure 6: Multi-joint abnormality detection for test video [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]
Figure 7
Figure 7. Figure 7: Degree of deviation from typical for test video [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]
Figure 8
Figure 8. Figure 8: Degree of deviation from typical for test video [PITH_FULL_IMAGE:figures/full_fig_p007_8.png]
Figure 9
Figure 9. Figure 9: Degree of deviation from typical for test video [PITH_FULL_IMAGE:figures/full_fig_p007_9.png]
Figure 10
Figure 10. Figure 10: Degree of deviation from typical for test video [PITH_FULL_IMAGE:figures/full_fig_p007_10.png]
Figure 11
Figure 11. Figure 11: Sequence of frames showing visualisation of potential gait abnormality in example with [PITH_FULL_IMAGE:figures/full_fig_p007_11.png]
Figure 12
Figure 12. Figure 12: Sequence of frames showing visualisation of potential gait abnormality in example with [PITH_FULL_IMAGE:figures/full_fig_p008_12.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

43 extracted references · 41 canonical work pages

  1. [1]

    Latest trends in gait analysis using deep learning techniques: A systematic review,

    Dimple Sethi, Chandra Prakash, and Sourabh Bharti, “Latest trends in gait analysis using deep learning techniques: A systematic review,” in Artificial Intelligence and Speech Technology: Third International Conference, AIST 2021, Delhi, India, November 12–13, 2021, Revised Selected Papers. Springer, 2022, pp. 363–375

    work page 2021

  2. [2]

    Human action recognition and prediction: A survey,

    Yu Kong and Yun Fu, “Human action recognition and prediction: A survey,” International Journal of Computer Vision , vol. 130, no. 5, pp. 1366–1401, 2022

    work page 2022

  3. [3]

    Deep gait recognition: A survey,

    Alireza Sepas-Moghaddam and Ali Etemad, “Deep gait recognition: A survey,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 45, no. 1, pp. 264–284, 2023

    work page 2023

  4. [4]

    Detection of Gait Abnormalities caused by Neurological Disorders,

    Daksh Goyal, Koteswar Rao Jerripothula, and Ankush Mittal, “Detection of Gait Abnormalities caused by Neurological Disorders,” in 2020 IEEE 22nd International Workshop on Multimedia Signal Processing (MMSP), Sept. 2020, pp. 1–6, ISSN: 2473-3628

    work page 2020

  5. [5]

    Biomechanical parameters for gait analysis: a systematic review of healthy human gait,

    Mary Roberts, David Mongeon, and Francois Prince, “Biomechanical parameters for gait analysis: a systematic review of healthy human gait,” Phys. Ther. Rehabil, vol. 4, no. 6, 2017

    work page 2017

  6. [6]

    Gait analysis methods in rehabilitation,

    Richard Baker, “Gait analysis methods in rehabilitation,” Journal of neuroengineering and rehabilitation, vol. 3, pp. 1–10, 2006

    work page 2006

  7. [7]

    Hybrid deep neural network framework combining skeleton and gait features for pathological gait recognition,

    Kooksung Jun, Keunhan Lee, Sanghyub Lee, Hwanho Lee, and Mun Sang Kim, “Hybrid deep neural network framework combining skeleton and gait features for pathological gait recognition,” Bioengi- neering, vol. 10, no. 10, pp. 1133, 2023

    work page 2023

  8. [8]

    Michael W Whittle, Gait analysis: an introduction , Butterworth- Heinemann, 2014

    work page 2014

Show all 43 references
  1. [9]

    The role of gait analysis in treating gait abnormalities in cerebral palsy,

    Frank M Chang, Jason T Rhodes, Katherine M Flynn, and James J Carollo, “The role of gait analysis in treating gait abnormalities in cerebral palsy,” Orthopedic Clinics, vol. 41, no. 4, pp. 489–506, 2010

    2010

  2. [10]

    Gait disturbances in old age: classification, diagnosis, and treatment from a neurological perspective,

    Klaus Jahn, Andreas Zwergal, and Roman Schniepp, “Gait disturbances in old age: classification, diagnosis, and treatment from a neurological perspective,” Deutsches ¨Arzteblatt International , vol. 107, no. 17, pp. 306, 2010

    2010

  3. [11]

    Psychosocial aspects of disorders affecting balance and gait,

    Lucy Yardley, Marjan Jahanshahi, and Richard S Hallam, “Psychosocial aspects of disorders affecting balance and gait,” Clinical disorders of balance, posture and gait , vol. 2, pp. 360–381, 2004

    2004

  4. [12]

    Gait, lower extremity strength, and self-assessed mobility after hip arthroplasty,

    Lola Sicard-Rosenbaum, Kathye E Light, and Andrea L Behrman, “Gait, lower extremity strength, and self-assessed mobility after hip arthroplasty,” The Journals of Gerontology Series A: Biological Sciences and Medical Sciences , vol. 57, no. 1, pp. M47–M51, 2002

    2002

  5. [13]

    The impact of ankle-foot orthosis stiffness on gait: A systematic literature review,

    Deema Totah, Meghna Menon, Carlie Jones-Hershinow, Kira Barton, and Deanna H Gates, “The impact of ankle-foot orthosis stiffness on gait: A systematic literature review,” Gait & posture , vol. 69, pp. 101– 111, 2019

    2019

  6. [14]

    Efficacy of clinical gait analysis: A systematic review,

    Tishya AL Wren, George E Gorton III, Sylvia Ounpuu, and Carole A Tucker, “Efficacy of clinical gait analysis: A systematic review,” Gait & posture, vol. 34, no. 2, pp. 149–153, 2011. Fig. 7: Degree of deviation from typical for test video one from GPJATK dataset [33] with typi...

    2011

  7. [15]

    Gait analysis in parkinson’s disease: An overview of the most accurate markers for diagnosis and symptoms monitoring,

    Lazzaro Di Biase, Alessandro Di Santo, Maria Letizia Caminiti, Alfredo De Liso, Syed Ahmar Shah, Lorenzo Ricci, and Vincenzo Di Lazzaro, “Gait analysis in parkinson’s disease: An overview of the most accurate markers for diagnosis and symptoms monitoring,” Sensors, vol. 20, no...

    2020

  8. [16]

    Reliability of observational kinematic gait analysis,

    David E Krebs, Joan E Edelstein, and Sidney Fishman, “Reliability of observational kinematic gait analysis,” Physical Therapy, vol. 65, no. 7, pp. 1027–1033, 1985. Frame 1 Frame 10 Frame 30 Fig. 12: Sequence of frames showing visualisation of potential gait abnormality in exam...

    1985

  9. [17]

    Use of videotape to assess mobility in a controlled randomized crossover trial of physiotherapy in chronic multiple sclerosis,

    CM Wiles, RG Newcombe, KJ Fuller, A Jones, and M Price, “Use of videotape to assess mobility in a controlled randomized crossover trial of physiotherapy in chronic multiple sclerosis,” Clinical rehabilitation, vol. 17, no. 3, pp. 256–263, 2003

    2003

  10. [18]

    Evalu- ation of 3D Markerless Motion Capture Accuracy Using OpenPose With Multiple Video Cameras,

    Nobuyasu Nakano, Tetsuro Sakura, Kazuhiro Ueda, Leon Omura, Arata Kimura, Yoichi Iino, Senshi Fukashiro, and Shinsuke Yoshioka, “Evalu- ation of 3D Markerless Motion Capture Accuracy Using OpenPose With Multiple Video Cameras,” Frontiers in Sports and Active Living , vol. 2, p...

    2020

  11. [19]

    Computer vision’s potential to improve health care,

    Arnold Milstein and Eric J. Topol, “Computer vision’s potential to improve health care,” The Lancet, vol. 395, no. 10236, pp. 1537, May 2020, Publisher: Elsevier

    2020

  12. [20]

    Automatic real- time gait event detection in children using deep neural networks,

    Łukasz Kidzi ´nski, Scott Delp, and Michael Schwartz, “Automatic real- time gait event detection in children using deep neural networks,” PloS one, vol. 14, no. 1, pp. e0211466, 2019

    2019

  13. [21]

    Deep neural networks enable quantitative movement analysis using single-camera videos,

    Łukasz Kidzi ´nski, Bryan Yang, Jennifer L Hicks, Apoorva Rajagopal, Scott L Delp, and Michael H Schwartz, “Deep neural networks enable quantitative movement analysis using single-camera videos,” Nature communications, vol. 11, no. 1, pp. 4054, 2020

    2020

  14. [22]

    Normal and pathological gait classification lstm model,

    Margarita Khokhlova, Cyrille Migniot, Alexey Morozov, Olga Sushkova, and Albert Dipanda, “Normal and pathological gait classification lstm model,” Artificial intelligence in medicine , vol. 94, pp. 54–66, 2019

    2019

  15. [23]

    A deep-learning approach for automatically detecting gait-events based on foot-marker kinematics in children with cerebral palsy-Which markers work best for which gait patterns?,

    Yong Kuk Kim, Rosa M. S. Visscher, Elke Viehweger, Navrag B. Singh, William R. Taylor, and Florian V ogl, “A deep-learning approach for automatically detecting gait-events based on foot-marker kinematics in children with cerebral palsy-Which markers work best for which gait pa...

    2022

  16. [24]

    Skeleton-based abnormal gait recognition with spatio-temporal attention enhanced gait- structural graph convolutional networks,

    Haoyu Tian, Xin Ma, Hanbo Wu, and Yibin Li, “Skeleton-based abnormal gait recognition with spatio-temporal attention enhanced gait- structural graph convolutional networks,” Neurocomputing, vol. 473, pp. 116–126, Feb. 2022

    2022

  17. [25]

    Pathological-Gait Recognition Using Spatiotemporal Graph Convolu- tional Networks and Attention Model,

    Jungi Kim, Haneol Seo, Muhammad Tahir Naseem, and Chan-Su Lee, “Pathological-Gait Recognition Using Spatiotemporal Graph Convolu- tional Networks and Attention Model,” Sensors (Basel, Switzerland) , vol. 22, no. 13, pp. 4863, June 2022

    2022

  18. [26]

    Two-dimensional video-based analysis of human gait using pose estimation,

    Jan Stenum, Cristina Rossi, and Ryan T. Roemmich, “Two-dimensional video-based analysis of human gait using pose estimation,” PLOS Computational Biology , vol. 17, no. 4, pp. e1008935, Apr. 2021, Publisher: Public Library of Science

    2021

  19. [27]

    Multi-speed transformer network for neurodegenerative disease assessment and activity recognition,

    Mohamed Cheriet, Vincenzo Dentamaro, Mohammed Hamdan, Donato Impedovo, and Giuseppe Pirlo, “Multi-speed transformer network for neurodegenerative disease assessment and activity recognition,” Com- puter Methods and Programs in Biomedicine , vol. 230, pp. 107344, 2023

    2023

  20. [28]

    A Survey of Human Gait-Based Artificial Intelligence Applications,

    Elsa J. Harris, I-Hung Khoo, and Emel Demircan, “A Survey of Human Gait-Based Artificial Intelligence Applications,” Frontiers in Robotics and AI, vol. 8, pp. 749274, Jan. 2022

    2022

  21. [29]

    Gait Recognition Based on Deep Learning: A Survey,

    Claudio Filipi Gonc ¸alves dos Santos, Diego de Souza Oliveira, Leandro A. Passos, Rafael Gonc ¸alves Pires, Daniel Felipe Silva Santos, Lucas Pascotti Valem, Thierry P. Moreira, Marcos Cleison S. Santana, Mateus Roder, Jo Paulo Papa, and Danilo Colombo, “Gait Recognition Base...

    2023

  22. [30]

    Multi-task learning for gait-based identity recognition and emotion recognition using attention enhanced temporal graph convolutional network,

    Weijie Sheng and Xinde Li, “Multi-task learning for gait-based identity recognition and emotion recognition using attention enhanced temporal graph convolutional network,” Pattern Recognition , vol. 114, pp. 107868, 2021

    2021

  23. [31]

    STEP: Spatial Temporal Graph Convolutional Networks for Emotion Perception from Gaits,

    Uttaran Bhattacharya, Trisha Mittal, Rohan Chandra, Tanmay Randha- vane, Aniket Bera, and Dinesh Manocha, “STEP: Spatial Temporal Graph Convolutional Networks for Emotion Perception from Gaits,” Proceedings of the AAAI Conference on Artificial Intelligence , vol. 34, no. 02, p...

    2020

  24. [32]

    Computer vision for clinical gait analysis: A gait abnor- mality video dataset,

    Rahm Ranjan, David Ahmedt-Aristizabal, Mohammad Ali Armin, and Juno Kim, “Computer vision for clinical gait analysis: A gait abnor- mality video dataset,” arXiv preprint arXiv:2407.04190 , 2024

    2024

  25. [33]

    Calibrated and synchronized multi-view video and motion capture dataset for evaluation of gait recognition,

    Bogdan Kwolek, Agnieszka Michalczuk, Tomasz Krzeszowski, Adam Switonski, Henryk Josinski, and Konrad Wojciechowski, “Calibrated and synchronized multi-view video and motion capture dataset for evaluation of gait recognition,” Multimedia Tools and Applications, vol. 78, no. 22,...

    2019

  26. [34]

    A review of observational gait assessment in clinical practice,

    B. Toro, C. Nester, and P. Farren, “A review of observational gait assessment in clinical practice,” Physiotherapy Theory and Practice , vol. 19, no. 3, pp. 137–149, Jan. 2003

    2003

  27. [35]

    Perceiving amputee gait from biological motion: kinematics cues and effect of experience level,

    I.-Chieh Lee, Matheus M. Pacheco, Michael D. Lewek, and He Huang, “Perceiving amputee gait from biological motion: kinematics cues and effect of experience level,” Scientific Reports, vol. 10, pp. 17093, Oct. 2020

    2020

  28. [36]

    Labelbox: The best way to create and manage training data. software, labelbox,

    M Sharma, D Rasmuson, B Rieger, D Kjelkerud, et al., “Labelbox: The best way to create and manage training data. software, labelbox,” Inc, https://www. labelbox. com , 2019

    2019

  29. [37]

    Blazepose: On-device real- time body pose tracking,

    Valentin Bazarevsky, Ivan Grishchenko, Karthik Raveendran, Tyler Zhu, Fan Zhang, and Matthias Grundmann, “Blazepose: On-device real- time body pose tracking,” in CVPR Workshop on Computer Vision for Augmented and Virtual Reality . IEEE, 2020

    2020

  30. [38]

    Validation of markerless 3d human pose estimation using blazepose for gait analysis,

    Sabyasachi Chakraborty, Abhishek Nandy, Tanaka Yamaguchi, Vincent Bonnet, and Gentiane Venture, “Validation of markerless 3d human pose estimation using blazepose for gait analysis,” Sensors, vol. 21, no. 21, pp. 7315, 2021

    2021

  31. [39]

    Real-time 3d human pose estimation from video with applications to exergam- ing,

    Matteo Capecci, Maria Gabriella Ceravolo, Francesco Ferracuti, Sabrina Iarlori, Gianluca Monte, Fiora Pirri, and Emanuele Scalona, “Real-time 3d human pose estimation from video with applications to exergam- ing,” in 2021 IEEE International Conference on Consumer Electronics (...

    2021

  32. [40]

    Sports2d - angles from video,

    David Pagnon, “Sports2d - angles from video,” 2023

    2023

  33. [41]

    27 of Applied Mathematical Sciences, Springer-Verlag, New York, 01 1978

    Carl de Boor, A Practical Guide to Splines , vol. 27 of Applied Mathematical Sciences, Springer-Verlag, New York, 01 1978

    1978

  34. [42]

    Scikit-learn: Machine learning in Python,

    F. Pedregosa, G. Varoquaux, A. Gramfort, V . Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V . Dubourg, J. Vander- plas, A. Passos, D. Cournapeau, M. Brucher, M. Perrot, and E. Duch- esnay, “Scikit-learn: Machine learning in Python,” Journal of Machine ...

    2011

  35. [43]

    1–19, Springer International Publishing, 05 2017

    Richard Baker, Fabien Leboeuf, Julie Reay, and Morgan Sangeux, The Conventional Gait Model : The Success and Limitations , pp. 1–19, Springer International Publishing, 05 2017

    2017

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.