Research on Unmanned Aerial Vehicle (UAV) Visual Landing Guidance and Positioning Algorithms
Jingping Shi
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThis paper is interesting and well written, the contributions are clearly introduced and supported by the experimental results in three lines: runway line detection and visual positioning system, optimization of image processing including the loss function and feature extraction steps, and application of Kalman filter to fuse the IMU and the visual positioning information.
Some recommendations to improve its readibility and focus:
-The text is not a thesis but a journal paper, so "this thesis" must be replaced by "this paper"
-The introduction describes several alternatives (runway, cooperative markers, deep learning) without clear connection among them, maybe the overall picture could be explained first to relate the relation among different blocks
-The calculus for vehicle localization based on the detected runway is an interesting and complex task, but more details should be included to explain this triangulation process. It is needed to see where the context information about airport runway is used to obtain the relative coordinate transform from runway detected in the image to vehicle coordinates.
-The justification for using neural detectors versus classical techniques such as the Hough transform is well justified, but the justification of adaptive filter versus the traditional Kalman filter is not clearly showed, it could be done in terms of the error in the estimation in the images depending on the distance.
Besides, it is remarked that as the number of sensors increases, the accuracy and robustness of the navigation system continuously improve, enhancing the performance of autonomous UAV landing. However, robust fusion solutions are need with the sources correctly characterized and systematic errors removed to guarantee this continuous improvement. It is indicated that Kalman filter is used to smooth the spatial trajectory, approximating the area where the target is likely to appear in the current frame. Although smoothing is an important aspect, the systematic errors must be characterized and removed, see for instance the work:
J.P. Llerena, J. García, J.M. Molina. Error Reduction in Vision-Based Multirotor Landing System. Sensors 2022
Wubben J., Fabra F., Calafate C.T., Krzeszowski T., Marquez-Barja J.M., Cano J.C. and Manzoni P. "Accurate Landing of Unmanned Aerial Vehicles Using Ground Pattern Recognition" Electronics 2019, 8(12), 1532
-The use of real data after simulation is an appealling aspect, although fell quite short in experiments, for example with changes in brightness due to the time of day or even certain occlusions on the runway lines. Besides, the evaluation metric with the real data is not clearly explained, and how the fusion with IMU can improve the results
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsIn the paper, the authors try to solve visual navigation problem in UAVs and they proposes a new approach to UAV visual landing guidance and positioning. Their approach uses the self-developed coarse positioning and deep learning-based runway line detection algorithms.
Suggestions:
The more extended explanations to figure titles should be given. Some sentence formulation needs to be improved, as example:
This thesis simulated the landing process of the UAV ..., - it should be changed to:
In the work (reserach), the landing process of the UAV on a simplified runway, treating the left and right edges of the runway as the left and right runway lines are simulated, respectively.
this thesis also calculated the error characteristics... - In the work, the error characteristics of the localization results in the three directions is also calculated..
The word "thesis" should be changed to "paper" , "work" or "research".
The conclusion should be improved. The figures of experiment results should be moved to the paper part "Experiment results".
In the conclusion, the detection accuracy and real-time performance advantages, generalization ability of the proposed algorithms over existings traditional UAV visual landing navigation algorithms and approaches should be quantitatively given.
Comments on the Quality of English Language
After authors' correction, the paper can be accepted for pulblication in the journal "Drones".
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsThe article proposes a deep learning based unmanned aerial vehicle (UAV) vision guided landing localization algorithm, which combines YOLOX and deep learning methods to achieve runway line detection and visual/IMU fusion localization. The algorithm design, implementation, and experimental verification are carried out, demonstrating the complete research process with strong innovation, practicality, and completeness. The significance of the topic selection in the paper is clear, the idea is relatively new, and it has strong theoretical innovation and engineering application value. Therefore, this paper meets the publishing requirements of drones. However, there are still some minor issues and omissions in the paper that need further revision by the author. My suggestion is to make minor revisions.
Point 1:. Suggest explaining the parameters (AP0.75 F1 Recall (%) Precision (%) Flops (G) Param (M)) in Chapter 2 and Chapter 3 evaluation tables.
Point 2. There is a line break in the English column in Figure 5. Please adjust it and check the other English formats and spelling in the figure.
Point 3. The expression of runway lines in the article is inconsistent, with some parts being track lines and others being runway lines. It is recommended to unify.
Point 4. What are the sources of errors in positioning algorithms? Please analyze and explain.
Point 5. The author may consider incorporating future research work into the conclusion of the paper at their discretion.
Point 6. The abstract needs further revision, reducing length and improving organization, while paying attention to the accuracy of wording in English abstracts;
Point 7. The non scalar symbols in the formula of the visual positioning algorithm in Chapter 5 should be bolded, please modify them;
Point 8. The positioning error evaluation table in Chapter 6 lacks units;
Point9. How to determine the variance matrix of system noise and measurement noise when applying Kalman filtering for navigation and positioning?
Comments on the Quality of English LanguageThe structure is reasonable and logical, and the design principle is clearly and moderately illustrated. The overall readability of the paper is good, and some sentences require minor modifications.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 4 Report
Comments and Suggestions for AuthorsThis manuscript proposes a UAV visual landing positioning system based on deep learning, with four main algorithms of runway coarse localization, runway line detection, visual positioning, and visual/inertial navigation. The simulation on customed datasets and experiment by actual flight verify the effectiveness of these components.
Some specific comments are listed below.
(1) There is too much content in the paper. As mentioned in the article, this is a ‘thesis’, not a research paper.
(2) In Section 2, the main objective is not clear. It does not make sense to study algorithms just for the sake of studying them. The application background and requirements should be analyzed to guide the study.
(3) In Section 2, the statement of ‘UAV pose estimation’ is not accurate. Pose includes position and attitude. The visual positioning algorithm is designed for estimating only position, while the attitude information is obtained from the IMU.
(4) The runway line detection method is complicated. Why cannot the runway line be directly segmented from the original image, but firstly perform coarse runway localization and then the runway line detection?
(5) In Section 3, many different loss functions, networks and components are compared and ablation studied. But it is confusing that which components are actually used in the proposed algorithm.
(6) In Section 5, the runway coordinate system is not defined. The runway line segmentation results include left and right lines, as well as starting lines, so which one is finally used for UAV position calculation?
(7) The real-time performance is not clearly demonstrated by the experimental results.
(8) It is chaotic and incoherent to place experimental results in theoretical subsections.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 5 Report
Comments and Suggestions for AuthorsThe article solves the complex problem of automated UAV landing. The authors consider several stages of landing, which include a visual analysis of the runway pattern and highlighting the runway boundaries.
The article contains several different topics, united by the topic of UAV landing control, so I believe it would be better to split it into several separate publications, which would significantly improve the readability of this material. Due to the large amount of work and the omission of details of the work, I am not sure that its results can be reproduced. At present, the work resembles a large technical report on a major scientific project, the presentation of which requires omitting a large number of details. This is definitely very interesting material.
The work sets out in great detail the essence of the matter and the set of problems that the authors had to solve. If the editorial policy allows the publication of articles of such a large volume, then the work as a whole can be published in its present form.
There are only two comments:
1. Authors use the word “thesis” (instead of “paper”?)
2. The drawings and captions for the axes in the drawings could be made larger. The text in the printed images is very difficult to read.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 4 Report
Comments and Suggestions for AuthorsThe author has made a reply to the review comments and made corresponding modifications and supplements in the revised manuscript. In my opinion, the revised manuscript has reached the academic level of this journal and is acceptable.
