Review Reports - Real-Time Hand Gesture Recognition for IoT Devices Using FMCW mmWave Radar and Continuous Wavelet Transform

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This paper presents an intelligent framework for real-time hand gesture recognition using FMCW mmWave radar and deep learning. Experiments show that the method is efficient and accurate intelligent in real-time IoT environments. The research topic in the paper is specific and interesting, the structure is reasonable, the expression is clear, the method is correct, and the conclusion is credible. Improvement is needed in the following areas to meet publication requirements:

1.Please add relevant research literature from the past 5 years and provide comments in the introduction section；

2.Compared to traditional methods, how much more computational cost will the method proposed in this article increase? Please provide an analysis of the experimental conclusions in the paper;

3.Please use appropriate punctuation marks in the formulas of the paper for ease of publication；

4.Please modify the figures in the paper to meet the requirements for publication;

5.What are the shortcomings and deficiencies of the method proposed in this article, and how should it be improved in the next step? It is recommended to provide comments before the conclusion of the paper and after the experiment.

Author Response

Dear Reviewer,

We sincerely thank you for your valuable feedback and constructive comments on our manuscript. We have carefully considered your suggestions and prepared detailed responses. Please find our replies compiled in the attached Word document.

Thank you for your time and consideration.

Best regards,
Anna Ślesicka

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The paper is well written and the topic is interesting. Nevertheless, I think there are some major concerns the Authors should address before its acceptance.

Main comments:

Consider the following articles, which can improve the state of the art and can be beneficial for the manuscript:
- Franceschini, S., Ambrosanio, M., Pascazio, V. and Baselice, F., 2022. Hand gesture signatures acquisition and processing by means of a novel ultrasound system. Bioengineering, 10(1), p.36.
- Franceschini, S., Ambrosanio, M., Baselice, F. and Pascazio, V., 2022, October. Person identification and authentication via ultrasound hand-gesture-signature analysis. In 2022 IEEE International Conference on E-health Networking, Application & Services (HealthCom) (pp. 229-233). IEEE.
Relatively to the results, provide confusion matrices and metrics beyond accuracy (e.g., precision/recall/F1).
Add details on augmentation, CNN architecture, and training settings, which have not been reported properly.
Improve statistical analysis by considering cross-validation, variance reporting, error bars, or similar analyses.
Clarify generalization limitations and discuss mitigation strategies.

Author Response

Dear Reviewer,

Thank you for your time and consideration.

Best regards,
Anna Ślesicka

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

1. The literature review mentions 62 types of gestures but lacks a unified standard. It is suggested to clarify the selection logic and specific basis of the standard gesture set, and refine the definition boundaries of different gestures.
2. Adopting CWT instead of traditional signal processing methods can supplement its direct comparison with STFT in terms of feature extraction performance and data processing time.
3. There were only 20 participants in the experiment, mostly family and friends, and the sample representativeness was insufficient. It is recommended to expand the sample size to include people of different age groups, hand sizes, and movement habits.
4. The in car experiment mentioned the presence of environmental interference, but did not specify the specific sources of interference (such as electronic devices, vehicle vibration, etc.) and control measures. It is recommended to provide a detailed analysis and elimination plan for the interference factors.
5. Data augmentation uses operations such as translation and rotation, but specific parameters such as translation pixels and rotation angle range are not specified. It is recommended to clarify these key settings.
6 When comparing different wavelet functions, only accuracy is considered without taking into account computational efficiency, while IoT devices have requirements for energy consumption and speed. It is recommended to supplement the comparison of computation time for each wavelet function.
7. Gestures only include changes in the index of reaching out, and the types are relatively single. It is recommended to add common interactive gestures such as sliding and grasping to enrich the application scenarios of research.
8. It is claimed that the model supports real-time recognition, but specific delay data is not provided. It is recommended to supplement the response time test results on different hardware devices to verify its real-time operation capability on IoT devices.
9. The accuracy rate of new users has dropped to 82-84%, and the reasons have not been analyzed in depth. It is recommended to explore the specific factors such as gesture execution speed and amplitude differences.
10. Select a 24GHz radar for research, without comparing the performance of other frequency bands (such as 60GHz, 77GHz). It is recommended to supplement the basis for frequency band selection.
11. In the experiment, each gesture was only repeated twice by the participants, which may result in accidental errors. It is recommended to increase the number of repetitions for a single gesture and record the consistency data of the action execution.
12. The study mentioned the future integration of multiple sensors, but did not provide a preliminary feasibility analysis. It is suggested to supplement the technical path concept or pre experimental results of fusion with sensors such as cameras and IMUs.

Author Response

Dear Reviewer,

Thank you for your time and consideration.

Best regards,
Anna Ślesicka

Author Response File: Author Response.pdf

Reviewer 4 Report

Comments and Suggestions for Authors

Why didn't the authors directly compare their approach with baselines based on FFT or STFT on the same dataset? The benefit of CWT would be more clearly demonstrated in a head-to-head comparison.
What is the difference between the suggested CNN architecture and more sophisticated models like CNN-LSTM hybrids or Transformers? Please consider or support your network selection.
How well does the model generalize to gestures with occlusion, overlapping motions, or quick execution? In practical IoT applications, these are typical.
Many models for gesture recognition are unable to generalize to users who were not present during training. This paper from Chen et al. provides insight into multi-modal collaboration with gesture recognition using deep learning architectures and can support the authors' cross-user evaluation by highlighting the significance of user-independent design: https://doi.org/10.1115/1.4054297.
How does the system respond to gestures made at various distances or angles? The range (45–90 cm) is mentioned, but there is no quantitative robustness analysis.