GPLVINS: Tightly Coupled GNSS-Visual-Inertial Fusion for Consistent State Estimation with Point and Line Features for Unmanned Aerial Vehicles

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The article entitled ‘GPLVINS: Tightly Coupled GNSS-Visual-Inertial Fusion for Consistent State Estimation with Point and Line Features for Unmanned Aerial Vehicles’ is clear and logical, and its structure complies with scientific publication standards. The text is coherent and fits well with current research on unmanned aerial vehicle navigation systems. Most of the sources cited are from the last five years and include key works in the field of sensor fusion and visual-inertial odometry, confirming the timeliness of the literature review. No excessive number of self-citations was found. The research methodology is scientifically sound – the stages of GNSS, IMU and image data processing, as well as the non-linear optimisation procedure, are described in detail. The level of detail in the Methods section allows for potential replication of the experiments. The results are presented in a transparent manner, and the tables and graphs are clear and relevant to the data discussed. The interpretation of the results is consistent and based on empirical evidence, with error analysis and comparison with previous solutions (GVINS) confirming the effectiveness of the proposed approach. The conclusions are logically linked to the presented results and do not go beyond the data obtained. The ethical and data availability statements are sufficient, although limited access to data due to confidentiality clauses may hinder full verification of the research. Overall, the manuscript is a valuable and well-developed contribution to the development of GNSS-visual-inertial fusion technology in UAV applications.

Comments that could significantly improve the quality of the scientific article ‘GPLVINS: Tightly Coupled GNSS-Visual-Inertial Fusion for Consistent State Estimation with Point and Line Features for Unmanned Aerial Vehicles’:

Expansion of the comparative analysis – it is worth including more reference methods (e.g. PL-VIO, PLS-VIO, OpenVINS) in the comparative experiments. Currently, the analysis is mainly limited to a comparison with GVINS, which narrows the scientific context.

Supplementing details on experiment parameters – in the ‘Experiment’ section, the system parameter settings (e.g., line detection threshold values, number of frames in the optimisation window, sensor frequency) should be clarified to facilitate replication of the results.

Strengthening the statistical analysis – the results could be supplemented with statistical analysis (e.g., standard deviation, RMSE significance test), which would increase the reliability of the comparisons.

Improvement of the quality of illustrations and figure descriptions – some graphs and photos (e.g. Figs. 5–9) are too small or described in an abbreviated manner. Adding clear legends, axes and unit markings would improve readability.

Discussion of the limitations of the proposed method – the ‘Conclusion’ section could include a brief analysis of the limitations of the GPLVINS system (e.g. dependence on GNSS signal quality, impact of low lighting, UAV computational limitations).

Expansion of the section on future work – it would be useful to specify in more detail the directions for further research, e.g. integration with real-time UAV control, testing in dynamic urban environments, or extension to multi-sensor systems with radars.

Author Response

Thank you for your valuable feedback and expert guidance! We have carefully revised the article in accordance with your suggestions. Please find the detailed revisions in the attached document for your further review.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

Revisão

• The authors should clearly define the acronym GPLVINS at its first mention in the abstract or introduction. For instance:
  “GPLVINS (GNSS-Point-Line-Visual-Inertial Navigation System) is a tightly coupled fusion framework designed for UAV state estimation.”

• The manuscript would greatly benefit from the inclusion of an Acronym List (either at the beginning or end of the article) to enhance readability and comprehension. The following acronyms, among others, should be defined:

  • LSD: Line Segment Detector

  • LBD: Line Band Descriptor

  • KNN: K-Nearest Neighbors

  • SVD: Singular Value Decomposition

  • VI: Visual-Inertial

  • ECEF: Earth-Centered, Earth-Fixed coordinate system

  • SPP: Single Point Positioning

  • RTK: Real-Time Kinematic

  • RMSE: Root Mean Square Error

  • ENU: East-North-Up coordinate system

  • EKF: Extended Kalman Filter

• In Equations (2) and (3), all variables, constants, and parameters should be explicitly defined or properly referenced immediately following their introduction. This includes terms such as ( r_D ), ( z̃_{sj}^{rk} ), ( X ), ( σ_{sj}^{rk,dp} ), ( δ̇t ), ( ψ ), ( r_P ), and ( r_T ). Readers unfamiliar with the GNSS estimation formulation might find these equations unclear without full definitions.

• The authors could also consider improving figure captions to be more descriptive. For example, captions such as “Effect diagram of line feature detection” could be expanded to specify dataset, parameters, and what improvements are visually evident.

• The Abstract and Introduction sections might be slightly restructured for clarity — explicitly stating the motivation, the problem in GVINS, and how GPLVINS addresses it. For example, highlight the novelty: “Unlike traditional GVINS methods that rely solely on point features, GPLVINS integrates line constraints to improve robustness in weak-texture environments.”

• It is recommended to include a brief discussion on computational efficiency, since the LSD algorithm and its optimization via NMS and length thresholding are central to the paper’s contribution. Quantitative results (e.g., reduction in processing time or computational cost) would strengthen the validation of the proposed approach.

• In the Results and Discussion section, the authors should ensure that figures comparing GPLVINS and GVINS trajectories are legible, with consistent color schemes and units. Including quantitative metrics such as runtime, frame rate, and memory usage could further support claims of UAV suitability.

• Finally, a short “Limitations and Future Work” paragraph could be added before the conclusion, summarizing current constraints (e.g., reliance on monocular vision, potential line extraction errors in highly dynamic scenes) and outlining next steps (e.g., extension to stereo or LiDAR-assisted systems).

Author Response

Thank you for your valuable feedback and expert guidance! We have carefully revised the article in accordance with your suggestions. Please find the detailed revisions in the attached document for your further review.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

There are two important points regarding GNSS fusion. I hope authors can emphasize them a bit more.

1. How to align Imu frame to GNSS frame? I mean to rotate Imu estimated heading to North? Can you highlight this part a bit more? Besides, is Imu required to mounted in some particular positions or can Imu be freely mounted? E.g a lot of car applications will have imu mounted in a way that x axis is aligned with the driving direction 
2. How do you deal with bad GNSS data? Do you have some weighting? How do you detect whether GNSS data is bad?

Author Response

Thank you for your valuable feedback and expert guidance! We have carefully revised the article in accordance with your suggestions. Please find the detailed revisions in the attached document for your further review.

Author Response File: Author Response.pdf

Reviewer 4 Report

Comments and Suggestions for Authors

This paper presents a tightly coupled GNSS–VI approach with point and line features for UAV localization. Authors claim to propose improvements with the LSD line extraction and NMS plus threshold filter. Nonetheless, this framework is too widely studied and the improvements seem to contribute with very little novelty.

Further contributions and additions to this work has to be done to demonstrate real advancement into the SOTA. 

Major concerns to be solved are:

- The paper mentions real-time feasibility for UAV onboard systems but lacks a quantitative analysis of computational cost, latency, or power consumption.

- Only GVINS is used as a baseline, comparisons with other recent VINS systems (e.g., PL-VIO, PLS-VIO, VINS-Fusion) are needed to provide a broader perspective of the novelty.

- The self-collected datasets are limited in scale (three indoor/outdoor scenarios) and lack statistical robustness. Larger-scale and public benchmarks (e.g., EuRoC, UZH-FPV) are needed to enhance generalizability.

- The contribution of the enhanced line extraction (NMS + threshold) versus standard LSD is qualitatively shown but not quantitatively isolated. Ablation study is needed.

- Some sections (especially GNSS initialization and optimization factors) are notation-heavy; simplifying explanations or providing symbolic legends would improve readability.

- The text could benefit from language polishing to enhance fluency and technical precision.

Overall, my main impression is that this work needs more comparison with other SOTA methods and generalized to wider datasets.

 

 

Author Response

Thank you for your valuable feedback and expert guidance! We have carefully revised the article in accordance with your suggestions. Please find the detailed revisions in the attached document for your further review.

Author Response File: Author Response.pdf

Reviewer 5 Report

Comments and Suggestions for Authors

This paper proposes GPLVINS, a UAV-oriented visual-internal navigation system that introduces point and line features based on GVINS for indoor GNSS-denied environments. GPLVINS is validated with open source datasets and self-collected UAV datasets in an indoor conference room, underground parking lot, and subway station. The results show that GPLVINS offers better positioning performance compared with the existing GVINS method. However, the reviewer cannot find any details about the system architecture, the setup of the real-world experiment and the positioning of the proposed method in relation to existing studies.

The reviewer's comments for this paper are as follows.

 

1. When using abbreviations in a paper, spell out the full term at first mention, followed by the abbreviation in parentheses. For example, GVINS, GNSS, LSD, and NMS do not have full terms.
2. This paper is an extension of GVINS, but the first time GVINS is used, no references are cited, which can confuse the readers. The authors should cite references where appropriate. In addition, the reviewer recommends that reference numbers be listed in order of citation.
3. List the contribution of the article at the end of the Introduction section.
4. Although Section 2 lists several publications, the differences between the proposed method and existing research are not clear. The authors should discuss the novelty of the proposed method.
5. Some of the references cited in the manuscript are relatively outdated, with many sources published before 2020. GVINS is a method proposed in a paper published in 2022, so there is a lack of up-to-date literature review.
6. The authors should provide details of the system model and the problem modeling in Section 3.
7. Some variables such as H and W in Equation (1) and P_{b_{t_k}}^w in Line 201 are not clearly defined.
8. The sizes of the conference room, underground parking and subway station scene shown in Figure 7 are not clear.
9. In Section 4.1, it is stated that ``Tab.1 shows that GPLVINS has lower RMSEs for latitude, longitude, and altitude data compared to GVINS”, but the differences between the results of GPLVINS and GVINS shown in Table 1 are minimal. Please add discussion of these numerical results and the limitations of GPLVINS.

Author Response

Thank you for your valuable feedback and expert guidance! We have carefully revised the article in accordance with your suggestions. Please find the detailed revisions in the attached document for your further review.

Author Response File: Author Response.pdf

Round 2

Reviewer 3 Report

Comments and Suggestions for Authors

The authors have addressed all my points and the paper is ready to publish. 

Author Response

Dear Reviewer,

Thank you very much for your rigorous review of our manuscript. Your insightful feedback throughout the review process has been invaluable in enhancing the quality and rigor of our work. We look forward to the final publication of the manuscript.

Sincerely,
Xinyu Chen
On behalf of all authors
Defense Innovation Institute, Chinese Academy of Military Science
E-mail: 657764423@qq.com

Reviewer 4 Report

Comments and Suggestions for Authors

My comments have been sufficiently addressed.

Author Response

Dear Reviewer,

Thank you very much for your rigorous review of our manuscript. Your insightful feedback throughout the review process has been invaluable in enhancing the quality and rigor of our work. We look forward to the final publication of the manuscript.

Sincerely,
Xinyu Chen
On behalf of all authors
Defense Innovation Institute, Chinese Academy of Military Science
E-mail: 657764423@qq.com

Reviewer 5 Report

Comments and Suggestions for Authors

The manuscript has been revised well and is in a nice condition now. However, I noticed some spelling inconsistencies in the revised manuscript (e.g., "optimization" vs. "optimisation", "minimize" vs. "minimisation" in Section 3.1). Please recheck your entire manuscript and revise it to ensure consistent spelling throughout.

 

there are no remaining issues.

Author Response

Comment1:
The manuscript has been revised well and is in a nice condition now. However, I noticed some spelling inconsistencies in the revised manuscript (e.g., "optimization" vs. "optimisation", "minimize" vs. "minimisation" in Section 3.1). Please recheck your entire manuscript and revise it to ensure consistent spelling throughout.
Response1: We have carefully addressed the comment regarding spelling inconsistencies. Specifically, we have unified the entire manuscript to follow the American English spelling system and conducted a comprehensive check to ensure no further inconsistencies. Key revisions include:

1. Standardizing all instances of British English spellings to American equivalents, replacing "minimisation" with "minimization" on page 4, section 3.1, line 147. Upon review, it was found that "optimization" and "utilization" were not spelled using British conventions in the text.
2. Conducting a full-text review covering the abstract, main text, figure captions, table contents, and references to confirm no mixed spelling of regional variants (e.g., "color" vs. "colour", "center" vs. "centre", "analyze" vs. "analyse").
3. Verifying the consistency of professional terms and abbreviations, ensuring uniform spelling throughout the manuscript without regional variant confusion