Deep Learning-Based Vehicle Speed Estimation Using Smartphone Sensors in GNSS-Denied Environment

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors 1. Generalizability: Data collection was limited to a single parking lot. Please discuss this limitation and consider evaluating the model in more diverse environments. 2.Reproducibility: The manuscript lacks details on code, data, and model availability. Please include a data/code availability statement. 3.Statistical Analysis: Report error variance or confidence intervals and apply statistical tests to support performance claims. 4.Limitations: The conclusion should briefly acknowledge potential limitations (e.g., device variability, model drift, sensor noise). 5.Comparative Baselines: Consider adding results or discussion on classical methods (e.g., Kalman filter) for context.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

This study presents a deep learning-based framework for vehicle speed estimation in GNSS-denied environments, using only smartphone sensors. Before it can be officially published, the following issues need to be further improved:

1. 1.The abstract does not mention the research background and issues, which need to be supplemented. In addition, the experimental conclusions mentioned in the abstract need to be further quantified to improve the credibility of the conclusions.
2. The design of some graphics and tables in this article is not standardized enough, which affects the expression of key data. For example, suggest adding standard deviation or interquartile range error lines in Figure 5-7, Table 2-4 has dense data and poor readability. Suggest using these tables as an appendix and placing it at the end of the main text.
3. 3.The literature review section of the paper is insufficient, with only 20 references. It is recommended that the author expand the scope of the literature review and cite more relevant research literature from the past 3 years, such as, https://org/10.3390/app15116255, https://doi.org/10.3390/jmse11112065, https://doi.org/10.1016/j.physa.2023.128980, https://doi.org/10.3390/app12062907.
4. In Section 3 of the paper, it is mentioned that the exclusion of the magnetometer is due to environmental interference, but no quantitative evidence is provided.
5. Why uses sample variance when calculating statistical features in Formula (7). It is recommended to supplement the reason for not selecting population variance.
6. Suggest adding a separate discussion section.
7. The conclusion lacks sufficient content, it is recommended that the authors further improved the conclusion to summarize the contributions from this study and supplement the analysis on the application of the research findings.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

Using sensor data from smartphones to determine a vehicle's speed is indeed a very interesting new topic worthy of further exploration.

However, unfortunately, the author does not explain how the raw data from the inertial sensors is converted into vehicle speed. Instead, the paper focuses extensively on the network architecture and experimental results, lacking a core explanation of the mapping function from raw data to speed, making it impossible for readers to replicate the experiments. While deep learning could potentially demonstrate a relationship between the raw data features and vehicle speed, directly providing speed values remains unconvincing.

From an experimental perspective, the limitations may stem from the parking lot scenario, where vehicle paths are overly regular. It is recommended to collect data on real roads to obtain more realistic complex vehicle motion data. Although outdoor scenarios do not represent a true GNSS-denied environment, such information can be excluded during data processing and instead used as ground truth for comparison.

Tables 2, 3, and 4 present comparisons of different experimental results, but the data is too voluminous to discern the information the authors aim to highlight. It is recommended to emphasize key data through bold formatting or other means.

Additionally, how does the algorithm perform in real-time? What is its computational resource consumption? Can it be directly applied to smartphone platforms?

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The authors have addressed all the reviewer comments appropriately. The limitations are now clearly discussed, additional statistical analyses have been added, and a data/code availability statement is included. The comparison with classical methods is helpful, and the paper is now well-rounded. I have no further concerns. I recommend acceptance.

 

Author Response

Thank you for your kind comments.

Reviewer 2 Report

Comments and Suggestions for Authors

Thank you to the author for their careful response to the previous round of review comments and efforts. However, after carefully reviewing the revised manuscript and the author's response, I believe that key issues in the manuscript have not been fully addressed.  The main reason is，I am not convinced by the author's response regarding opinions 3, 4, and 6 in the first opinion.  These questions are crucial for the scientific rigor, conclusion reliability, and theoretical contribution of the paper. I suggest that they be further improved.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

1. The reference 22 and 26 are duplicates..
2. In the discussion section, a comparison is given between the proposed algorithm and KF and DNN. KF is obtained by the IMU and GNSS data based on the physical model, but this paper is in a GNSS-denied environment and there is no GNSS data. This is obviously incorrect.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 3

Reviewer 2 Report

Comments and Suggestions for Authors

I believe that the current version of the manuscript has not yet met the publication standards of this journal.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf