Review Reports

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This manuscript presents an embedded, magnetometer-free inertial measurement unit (IMU) sensor system for the quantitative analysis of kendo movements. The study is timely and practically relevant, as it addresses a genuine need for objective movement assessment in kendo while preserving ease of use through sensor integration within the shinai. The combination of a compact embedded implementation, six-axis orientation estimation using an extended state Kalman filter (ESKF), gravity compensation, and peak-based movement feature extraction is technically sound and of clear interest. The article tackles an important applied problem in sports motion analysis, and the proposed embedded sensing approach represents a meaningful contribution. The experimental results are encouraging and suggest that the system can extract useful discriminative features from kendo movements; however, The practical and scientific added value is quite limited, and the hardware resources used are very basic and so several improvements are needed to strengthen the manuscript:
1- The paper does not provide a complete and reproducible description of the signal-processing pipeline. The authors should specify the full acquisition and analysis workflow, the exact preprocessing steps, the interpolation method used to obtain a uniform 20 Hz sampling rate, the motion and peak-detection criteria, and the implementation details of gravity compensation based on the ESKF. Although the manuscript states that the sampling frequency was set to 20 Hz and that linear interpolation was applied, the processing sequence remains insufficiently detailed to ensure reproducibility.
2- The authors should report the number of independent or repeated trial series conducted for each participant and clarify whether the reported statistics are based on a single trial, multiple trials, or pooled trials. Since the manuscript compares experienced and novice practitioners using peak acceleration, full width at half maximum (FWHM), and secondary peak ratio, it would be preferable to provide the mean ± standard deviation for all trials and, where appropriate, for all subjects. This is particularly important given the relatively small sample size (24 participants in total) and the fact that the conclusions rely on group-level comparisons.
3- The manuscript would benefit from a more systematic ablation study. While the current version highlights the efficiency of the final system, it would be valuable to quantify separately the contribution of each major design choice, including the integrated configuration, magnetometer-free estimation, the choice between ESKF and EKF, the gravity compensation step, and the peak-extraction strategy. Such an ablation table could also report the impact of each component on the main evaluation metrics, as well as any trade-offs in computational cost or implementation complexity.
4- The comparison with previous work should be made more precise. The manuscript cites several related systems for kendo movement analysis and emphasizes the novelty of integrated detection and magnetometer-free estimation; however, the reader would benefit from a clearer explanation of whether the compared methods were evaluated under identical conditions. When results are taken from previous publications, the authors should explicitly note differences in evaluation protocols and avoid direct numerical comparisons unless the conditions are truly comparable.
5- The quality and legibility of the figures should be improved. Several graphs would benefit from higher-resolution export and more readable fonts, especially since the paper relies heavily on waveform interpretation and visual comparison of peaks. The manuscript would be more convincing if the final version included sharper, high-resolution graphics.
6- The manuscript should undergo a thorough linguistic revision. A careful language edit would improve the readability and overall professionalism of the paper, particularly in the introduction, methodology, and discussion sections.
7- The bibliography is quite limited and should be expanded with additional references, especially more recent ones.

Comments on the Quality of English Language

The manuscript should undergo a thorough linguistic revision. A careful language edit would improve the readability and overall professionalism of the paper, particularly in the introduction, methodology, and discussion sections.

Author Response

Comments 1: 1- The paper does not provide a complete and reproducible description of the signal-processing pipeline. The authors should specify the full acquisition and analysis workflow, the exact preprocessing steps, the interpolation method used to obtain a uniform 20 Hz sampling rate, the motion and peak-detection criteria, and the implementation details of gravity compensation based on the ESKF. Although the manuscript states that the sampling frequency was set to 20 Hz and that linear interpolation was applied, the processing sequence remains insufficiently detailed to ensure reproducibility.
Response 1: We agree that a complete and clearly structured description of the signal-processing pipeline is essential for reproducibility. In the revised manuscript, we have provided a detailed description of all processing steps, including resampling using linear interpolation, smoothing, motion onset detection, peak detection, and feature extraction. In addition, we have explicitly clarified the overall processing workflow to improve clarity and reproducibility. The full signal-processing pipeline is now described as a sequence of steps, including data acquisition, resampling, orientation estimation using the ESKF, gravity compensation, and subsequent feature extraction. These revisions have been incorporated into Section 2.6.

Comments 2: 2- The authors should report the number of independent or repeated trial series conducted for each participant and clarify whether the reported statistics are based on a single trial, multiple trials, or pooled trials. Since the manuscript compares experienced and novice practitioners using peak acceleration, full width at half maximum (FWHM), and secondary peak ratio, it would be preferable to provide the mean ± standard deviation for all trials and, where appropriate, for all subjects. This is particularly important given the relatively small sample size (24 participants in total) and the fact that the conclusions rely on group-level comparisons.
Response 2: We have clarified that two trials were recorded for each participant and that the second trial was used for the main statistical analysis, except when it was unavailable due to sensor malfunction. In addition, the mean and standard deviation of peak acceleration, FWHM, and secondary peak ratio for each trial have been added to Tables S1 and S2 in the Supplementary Material. In addition, the selection of trials used for analysis has been slightly revised to ensure consistency in the application of the predefined selection criteria. As a result, minor changes in numerical values and figures were observed; however, the overall statistical trends and conclusions remain unchanged.

Comments 3: 3- The manuscript would benefit from a more systematic ablation study. While the current version highlights the efficiency of the final system, it would be valuable to quantify separately the contribution of each major design choice, including the integrated configuration, magnetometer-free estimation, the choice between ESKF and EKF, the gravity compensation step, and the peak-extraction strategy. Such an ablation table could also report the impact of each component on the main evaluation metrics, as well as any trade-offs in computational cost or implementation complexity.
Response 3: We agree that a systematic ablation study could provide additional insight into the contribution of each component. However, the primary objective of this study is to demonstrate the feasibility and effectiveness of the proposed integrated sensing system for kendo motion analysis under realistic conditions rather than to optimize or benchmark individual components in isolation. Performing a full ablation study would require controlled experimental conditions and additional data collection, which are beyond the scope of the present study. To address the reviewer's suggestion, we have added a discussion clarifying this limitation and highlighting the importance of future work to systematically evaluate the contribution of each component.

Comments 4: 4- The comparison with previous work should be made more precise. The manuscript cites several related systems for kendo movement analysis and emphasizes the novelty of integrated detection and magnetometer-free estimation; however, the reader would benefit from a clearer explanation of whether the compared methods were evaluated under identical conditions. When results are taken from previous publications, the authors should explicitly note differences in evaluation protocols and avoid direct numerical comparisons unless the conditions are truly comparable.
Response 4: In response, we have revised the manuscript to provide a more precise and structured comparison with related studies. Specifically, we have clarified that previous studies differ in terms of sensor placement, system configuration, experimental protocols, and research objectives. For each cited study, we now explicitly describe these differences, including whether the sensors were attached to the body or the shinai, whether the focus was on activity recognition or motion analysis, and whether the study addressed subjective experience or quantitative measurement. Furthermore, we have avoided direct numerical comparisons with previous studies, as the experimental conditions are not identical. Instead, the comparison has been reframed to highlight differences in methodology and application context. For example, previous IMU-based studies (e.g., Takata et al. and Torigoe et al.) focus on activity classification using machine learning, whereas the present study focuses on the extraction of motion characteristics and the quantitative evaluation of differences between experienced and novice practitioners. These clarifications have been incorporated into the revised manuscript (Section 1.2), and we believe that they improve the transparency and appropriateness of the comparison with prior work.

Comments 5: 5- The quality and legibility of the figures should be improved. Several graphs would benefit from higher-resolution export and more readable fonts, especially since the paper relies heavily on waveform interpretation and visual comparison of peaks. The manuscript would be more convincing if the final version included sharper, high-resolution graphics.
Response 5: We have revised the figures to improve their quality and legibility. The resolution of the figures has been increased, and the font sizes in the graphs have been enlarged to make waveform patterns, peak positions, and group comparisons easier to read.

Comments 6: 6- The manuscript should undergo a thorough linguistic revision. A careful language edit would improve the readability and overall professionalism of the paper, particularly in the introduction, methodology, and discussion sections.
Response 6: The manuscript has been carefully revised to improve clarity, readability, and overall language quality. In particular, the Introduction, Methods, and Discussion Sections have been refined to enhance coherence and improve the overall professionalism of the paper. In addition, the revised manuscript has undergone professional English proofreading to ensure consistency and correctness.

Comments 7: 7- The bibliography is quite limited and should be expanded with additional references, especially more recent ones.
Response 7: In response, we have expanded the bibliography by incorporating additional relevant references, particularly focusing on recent developments in IMU-based sensing and orientation estimation. Specifically, we have added representative studies such as Madgwick et al. (2011), which provides a widely used classical approach to IMU-based orientation estimation, and Laidig and Seel (2023), which reflects recent advances in this field. These additions complement the existing references and provide a more balanced coverage of both classical and modern approaches. While the number of additional references is limited, we have carefully selected works that are directly relevant to the scope of this study. These additions improve the contextualization of the proposed method within the current state of the literature.

Comments 8: Comments on the Quality of English Language The manuscript should undergo a thorough linguistic revision. A careful language edit would improve the readability and overall professionalism of the paper, particularly in the introduction, methodology, and discussion sections.
Response 8: The manuscript has been carefully revised to improve clarity, readability, and overall language quality. In particular, the Introduction, Methods, and Discussion Sections have been refined to enhance coherence and improve the overall professionalism of the paper. In addition, the revised manuscript has undergone professional English proofreading to ensure consistency and correctness.

Reviewer 2 Report

Comments and Suggestions for Authors

1. The core claim of this paper—"attitude estimation without a magnetometer"—is theoretically flawed. Theoretically, it is fundamentally impossible to estimate yaw attitude without a magnetometer. This reflects a misunderstanding on the part of the authors. Furthermore, this technical approach is already existing technology, and the authors have made no contribution in this regard.
2. Kendo swings are high-speed, dynamic movements, and the duration of the impact process is typically on the order of tens of milliseconds. A sampling rate of 20 Hz means a sampling interval of 50 ms between consecutive data points, which fails to satisfy the fundamental requirements of the Nyquist sampling theorem. Therefore, I consider the conclusions drawn by the authors to be scientifically unreliable.
3. The methodology of this paper is not replicable. Key processing techniques lack specific parameters, including: the window size of the smoothing filter, the saliency threshold for peak detection, and others.
4. The statistical methods used in this paper are inappropriate. The authors directly applied a parametric t-test with a small sample size (approximately 12 people per group) without conducting any normality test. This operation is risky and statistically unsound.

Author Response

Comment 1: The core claim of this paper—"attitude estimation without a magnetometer"—is theoretically flawed. Theoretically, it is fundamentally impossible to estimate yaw attitude without a magnetometer. This reflects a misunderstanding on the part of the authors. Furthermore, this technical approach is already existing technology, and the authors have made no contribution in this regard.
Response 1: We agree that absolute yaw orientation cannot be uniquely determined from six-axis IMU data alone without a magnetometer, and we acknowledge this as a fundamental limitation. In this study, however, the objective is not to estimate absolute yaw orientation but to obtain an orientation estimate sufficient for gravity compensation and motion analysis. As clarified in the revised manuscript (Section 2.5), yaw rotation does not affect the direction of gravity and therefore has limited influence on gravity compensation. Furthermore, as discussed in Section 4.2, the kendo swing motion analyzed in this study is dominated by pitch motion, and the yaw component is relatively small. Therefore, the influence of yaw estimation errors on the extracted motion features is limited in this application. Regarding novelty, we agree that the use of the ESKF for IMU-based orientation estimation is not new. The contribution of this study lies not in proposing a new estimation algorithm but rather in the design and implementation of an embedded IMU-based sensing system integrated into a shinai, as well as in demonstrating its effectiveness for quantitative analysis of kendo swing motion under realistic training conditions. In addition, we have revised the title of the manuscript to better reflect the actual contribution of this study, emphasizing the embedded sensing system and its application rather than the orientation estimation method itself. The manuscript has been revised accordingly to clarify the scope, limitations, and contributions of the study.

Comment 2: Kendo swings are high-speed, dynamic movements, and the duration of the impact process is typically on the order of tens of milliseconds. A sampling rate of 20 Hz means a sampling interval of 50 ms between consecutive data points, which fails to satisfy the fundamental requirements of the Nyquist sampling theorem. Therefore, I consider the conclusions drawn by the authors to be scientifically unreliable.
Response 2: We agree that a sampling frequency of 20 Hz is insufficient to capture high-frequency dynamics associated with the impact phase of kendo strikes, which occur on the order of tens of milliseconds. However, the objective of this study is not to analyze the detailed impact process but rather to characterize the overall swing motion and its temporal structure. The features analyzed in this study, including peak acceleration, peak width (FWHM), and secondary peak structure, reflect lower-frequency components of the motion associated with the periodic swing of the shinai. The typical duration of one swing cycle is approximately 0.5-1.0 s (corresponding to frequencies below ~2 Hz), and the peak width is on the order of 0.1-0.3 s (corresponding to frequency components below ~5-10 Hz). Therefore, the selected sampling frequency of 20 Hz satisfies the Nyquist criterion for the motion features of interest. We acknowledge that high-frequency components during impact may not be captured and may introduce some degree of aliasing. However, these components are temporally localized and contribute only a small portion of the overall signal; thus, their influence on the extracted low-frequency features is expected to be limited. Importantly, the experimental results consistently show statistically significant differences between experienced and novice practitioners, suggesting that the selected features capture meaningful motion characteristics even at the current sampling rate. To clarify this point, we have revised the manuscript to explicitly state the scope and limitations of the current sampling frequency.

Comment 3: The methodology of this paper is not replicable. Key processing techniques lack specific parameters, including: the window size of the smoothing filter, the saliency threshold for peak detection, and others.
Response 3: In the revised manuscript, we have explicitly defined all key parameters used in the analysis. Specifically, the smoothing of the acceleration norm is now described using a moving average filter with a window size of 3 samples. Motion onset detection is defined using a robust threshold based on the median plus 1.5 times the median absolute deviation (MAD) of the acceleration norm, with a requirement that the threshold be exceeded for at least 3 consecutive samples. For peak detection, we have clarified that the main peaks are identified as local maxima in the Z-axis acceleration signal that exceed a saliency threshold defined as max(noise scale, 1.0 m/s²), where the noise scale is computed as 1.4826 times the MAD of the Z-axis acceleration signal. In addition, a minimum peak interval of 0.7 s is imposed to suppress false detections. These details have been incorporated into the revised manuscript (Section 2.6).

Comment 4: The statistical methods used in this paper are inappropriate. The authors directly applied a parametric t-test with a small sample size (approximately 12 people per group) without conducting any normality test. This operation is risky and statistically unsound.
Response 4: We have revised the manuscript to include a normality assessment using the Shapiro-Wilk test for each variable and group. The results indicated that normality was generally satisfied for some variables (e.g., peak acceleration and the secondary peak ratio), while deviations from normality were observed for others (e.g., FWHM in the novice group). In addition to Welch's t-test, which was originally used because of its robustness to unequal variances, we have performed the non-parametric Mann-Whitney U test for all group comparisons. The results obtained using the Mann-Whitney U test are consistent with those obtained using Welch's t-test, with significant differences observed for peak acceleration and FWHM, and no significant difference for the secondary peak ratio. These analyses help mitigate the potential limitations associated with the relatively small sample size and support the robustness of the conclusions. These additional analyses have been incorporated into the revised manuscript (Sections 2.6 and 3.3-3.5).

Reviewer 3 Report

Comments and Suggestions for Authors

This research deals with wearable sensor systems that are applied for motion analysis in sports. IMU unit with microcontroller is used for motion analysis. The article solves the current issue.

The introduction describes the motivation for the creation of this research and development. Related work with similar topics is also listed.
The next section describes a measurement sensor system based on the low-cost IMU unit MPU9250 and the Raspberry Pi Zero W microcontroller.
To process the measured signals, the authors used an error state Kalman filter (ESKF) using only 6-axis IMU data without relying on magnetometer measurements.
Different participants with different experiences were used for the experiments.
The results of the experiments confirm the correctness of the methods used to solve this problem. The system can record the characteristics of kendo movements. This represents the most significant contribution of this work.

Comments:
1. The article is more descriptive in nature and looks more like a student thesis. The authors did not disclose any details of their solution and the methodology used, which should be a benefit of this work. There is a lack of a mathematical description of the methodology used in the algorithm.

2. For example, the algorithm for processing signals from the IMU unit could be interesting. If possible, I recommend providing it in the form of pseudocode, for example.

3. How was the measurement system verified and calibrated?
How is it possible to trust this system without proper verification using a suitably chosen method. If these results are to be used correctly, then this problem should also be appropriately solved.

4. It would also be interesting to compare the results with other conventional methods so that the benefit of this work is visible. However, the authors did not do this. From this point of view, this work seems to me to be of little benefit. I therefore recommend describing the individual procedures and methods in more detail and devoting more attention to the comparative analysis of the experimental results.

Author Response

Comments 1: 1. The article is more descriptive in nature and looks more like a student thesis. The authors did not disclose any details of their solution and the methodology used, which should be a benefit of this work. There is a lack of a mathematical description of the methodology used in the algorithm.
Response 1: We agree that a clear and rigorous description of the methodology is essential. In the revised manuscript, we have substantially expanded the methodological description in Section 2.5. Specifically, we added an explicit mathematical formulation of the ESKF-based orientation estimation, including the definition of the nominal and error states, the state transition model, the measurement model, and the covariance propagation. In addition, we clarified the quaternion-based representation of rotation and the formulation used for gravity estimation and compensation, as well as the adaptive modeling of the measurement noise.

Comments 2: 2. For example, the algorithm for processing signals from the IMU unit could be interesting. If possible, I recommend providing it in the form of pseudocode, for example.
Response 2: In the revised manuscript, we have included a Python-like pseudocode representation of the complete signal-processing pipeline in Section 2.6 (Listing 1). This pseudocode summarizes the entire workflow, including resampling, orientation estimation using the ESKF, gravity compensation, motion onset detection, peak detection, and feature extraction.

Comments 3: 3. How was the measurement system verified and calibrated? How is it possible to trust this system without proper verification using a suitably chosen method. If these results are to be used correctly, then this problem should also be appropriately solved.
Response 3: The IMU sensor used in this study was verified through a basic static calibration procedure prior to data collection. Specifically, the sensor was placed in known orientations, and it was confirmed that the measured acceleration corresponded to approximately 1 g along the expected axis. We acknowledge that more rigorous calibration procedures (e.g., full multi-axis calibration or dynamic validation) were not performed. However, the primary objective of this study is to compare relative motion characteristics between groups under consistent conditions rather than to obtain absolute measurements with high precision. Since the same sensor and processing pipeline were applied uniformly to all participants, the relative differences observed in the results are considered reliable for the purpose of this study. We have clarified this point in the revised manuscript.

Comments 4: 4. It would also be interesting to compare the results with other conventional methods so that the benefit of this work is visible. However, the authors did not do this. From this point of view, this work seems to me to be of little benefit. I therefore recommend describing the individual procedures and methods in more detail and devoting more attention to the comparative analysis of the experimental results.
Response 4: We agree that a clear comparison with existing approaches is important for understanding the benefit of the proposed method. However, direct numerical comparison with previous studies is not appropriate in this case because of substantial differences in experimental conditions, including sensor placement, system configuration, and analysis objectives. In response, we have revised the manuscript to provide a more explicit discussion of the relationship between the proposed method and existing approaches. Specifically, we have clarified that previous IMU-based studies primarily focus on activity recognition or classification using multiple sensors, whereas the present study focuses on extracting motion characteristics from swing dynamics using an embedded sensing system integrated into a shinai. Furthermore, we have added a discussion in Section 4.1 to highlight the practical advantages of the proposed system. In particular, we emphasize that the system enables the distinction between experienced and novice practitioners using a simplified and unobtrusive measurement configuration, which is suitable for real training environments. These revisions clarify the practical benefit of the proposed system and provide a more appropriate comparison with existing approaches without relying on direct numerical comparisons under different experimental conditions.

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The bibliography is still quite limited. I encourage the authors to expand it by adding more references, especially recent ones.

Author Response

Comment:
The bibliography is still quite limited. I encourage the authors to expand it by adding more references, especially recent ones.

Response:
Thank you very much for this valuable suggestion.
We expanded the bibliography and related work section by adding recent and relevant references on wearable sensing, IMU-based sports motion analysis, sports equipment sensing, combat sports sensing, motion-capture-based kendo analysis, and orientation estimation methods.
We believe that these additions strengthen the positioning of the present study within the broader context of sports wearable sensing and motion analysis research.

Reviewer 2 Report

Comments and Suggestions for Authors

The references in this paper are insufficient, falling far below the basic standard for an SCI paper. The author needs to supplement them.

Author Response

Comment:
The references in this paper are insufficient, falling far below the basic standard for an SCI paper. The author needs to supplement them.

Response:
Thank you for this important comment.
In response, we substantially expanded the related work and bibliography sections throughout the manuscript.
The revised manuscript now includes additional references on wearable sensing, IMU-based sports motion analysis, embedded sensing systems, sports biomechanics, swing analysis in golf/baseball/ice hockey, combat sports analysis, motion-capture-based kendo analysis, and orientation estimation methods.
We believe that these revisions strengthen the scholarly background and positioning of the manuscript.

Reviewer 3 Report

Comments and Suggestions for Authors

All my comments and ambiguities have been resolved and clarified in the article. The article is of much better quality and also has a scientific level. This makes it interesting for readers. I did not find any other ambiguities and errors in the article. The article is now ready for publication.

Author Response

Comment:
All my comments and ambiguities have been resolved and clarified in the article. The article is of much better quality and also has a scientific level. This makes it interesting for readers. I did not find any other ambiguities and errors in the article. The article is now ready for publication.

Response:
We sincerely thank the reviewer for the positive evaluation and encouraging comments.
We further revised the manuscript according to the additional suggestions from the Editor and other reviewers to improve the overall clarity and quality of the paper.