Multi-Drone 3D Building Reconstruction Method
Round 1
Reviewer 1 Report
As an independent reviewer, I see that this paper has no important mathematical theory to be published in this journal. The main problem is that the contribution of this paper is very low. The topic of this paper is different from the scopes of mathematics journal. The novelties of this paper are not clear and I think that it is better that this paper be rejected. According to my comments to the paper, I recommend rejection of this paper.
Author Response
Hello, thank you for your review. We agree that this paper brings a little mathematical theory. However there exist several formulas that might be used to set up the described algorithms. Moreover the paper is submitted to the special issue "Application of Mathematical Methods in Artificial Intelligence". We suppose that the method and the formulas, that are described in this paper, are acceptable for the “Artificial Intelligence” topic.
Reviewer 2 Report
There are numerous spelling errors, s/v errors, missing punctuations, article errors that seriously affect reading and comprehending the paper.
Just to mention a few in the abstract!
line
1 in recent decad missing punctuation
3 works lacks s/v error
6 with RGB camera article error
7 perform s/v error
12 number of drones article error
17 perfoem spelling error
18 capturd spelling error
Proofreading must be done before this paper can be re-evaluated!
Section I introduction can be reorganized. Suggest avoiding using "conclusions" in the introduction.
Author Response
Hello, thank you for your review!
We have done the proofreading, thanks for pointing our attention.
We also have reorganized the introduction, so that conclusions are not mentioned there.
Reviewer 3 Report
This topic is very interesting and belongs to the applied mathematics however in the practice we use drones very commonly to create 3D modells for buildings with only moderate difficulties. Reviewer thinks this manuscript could fit even to other jourmnals as well as Photogrammetry or Drones (MDPI). Concept of the manuscript is clear and understandable, methods are also acceptable. Chapters and the flow of the paper is logic, results are also acceptable. I did not find mistake in formulas. Even if this paper is very interesting and reviewer found it nice its novelty evaluated moderate.
Author Response
Thank you for your review!
Point 1. Reviewer thinks this manuscript could fit even to other journals as well as Photogrammetry or Drones (MDPI)
Response: Thank you for the suggestion, we will consider it.
Point 2. Even if this paper is very interesting and reviewer found it nice its novelty evaluated moderate.
Response: We have added a couple of words in the introduction that describe the novelty of this paper from our point of view.
Reviewer 4 Report
The possibility of automatic flights for the extraction of 3D models and orthophotography is a challenge of great interest. Not only because of the quality of the products, but also because of the possibilities of safety and precision in flight that can be achieved. In this sense, the work presented is of great interest.
With the exception of certain improvements listed in the specific comments section, I would like to make some basic comments:
There are several problems inherent to these options, GPS positioning close to a building makes guidance complex and for this reason it seems that the visual odometer solution provided is promising. Being 5m from the object with a positioning error of 0.4m and with a trajectory error of 0.7m may seem low, but 20 percent of difference at that distance to the object is not negligible. The use of a 40 percent overlap seems low because with such errors the performance seems to be compromised. Many authors appreciate the need for at least 3 images per point and two (stereoscopy) to achieve a correct solution in Sfm methods. These parameters indicated as a result in 4.2 are indispensable and offer reasonable doubts to the article.
Don't the authors believe that this result may force to include it as a case of the present study?
Section 4.1 being a case of theoretical simulation of formulas and 4.2 a synthetic case, the authors should make the same case as in 4.2 but with the overlaps and number of frames per point to establish the comparison that they evaluate as a conclusion of results.
Image 10 doesn’t show the positions, distances and overlaps indicated in the realization of case 4.2. The results given in the article differ greatly from those shown in the image. The accuracy of 0.4m is not visible in the image, where much larger errors and great difficulties in the execution of the flight are visible. Can you include an image that allows us to observe compliance with this data?
The reviewer believes that including the case study obtained in the results and the modification of figure 10 should be included in the article.
Congratulations on the article and the work done.
Kind regards,
Specific comments:
1.- The description of acronyms such as RGBD, VSLAM, ROS, RTAB , FAST or FRCNN is missing.
2.- In 149 the sSfm algorithm is mentioned, the acronym should be indicated and, if possible, a minimal explanation that allows its inclusion in the article to be understood.
3.- The concept of heterogeneous architecture is used on three occasions, you can explain it on one of the three occasions.
4.- It is necessary to revise, in addition to the English, words that are misspelled such as: perfoem (17), intial (257), recieving (260), please revise.
5.- Tables 2 and 3 should be joined so that the headers can be read and interpreted, in the caution you can indicate the units and make some reference if you cannot include information in the header.
Author Response
Hello, thank you for your detailed review! Please find our responses point by point below.
Point 1. There are several problems inherent to these options, GPS positioning close to a building makes guidance complex and for this reason it seems that the visual odometer solution provided is promising. Being 5m from the object with a positioning error of 0.4m and with a trajectory error of 0.7m may seem low, but 20 percent of difference at that distance to the object is not negligible. The use of a 40 percent overlap seems low because with such errors the performance seems to be compromised. Many authors appreciate the need for at least 3 images per point and two (stereoscopy) to achieve a correct solution in Sfm methods. These parameters indicated as a result in 4.2 are indispensable and offer reasonable doubts to the article.
Don't the authors believe that this result may force to include it as a case of the present study?
Response: We agree that the error in 20% of the distance between a drone and a building is considerable. This result is obtained without using SLAM techniques and it is expected that applying SLAM algorithms will improve the accuracy of positioning. Moreover the results, that are presented in table 2 from section 4, may be used to improve this quality.
Point 2. Section 4.1 being a case of theoretical simulation of formulas and 4.2 a synthetic case, the authors should make the same case as in 4.2 but with the overlaps and number of frames per point to establish the comparison that they evaluate as a conclusion of results.
Response: We added the calculations for the case from section 4.2 to section 4.1. The main goal of theoretical section 4.1 is to present formulas that connect parameters of the algorithm. Unfortunately these parameters cannot be connected to the resulting accuracy of 3D reconstruction. The closest thing that can be evaluated is the percentage of images overlay. The results of the experimental evaluation in section 4.2 fit the expectation, and you mentioned correctly, that 40% of images overlay is not enough for accurate reconstruction.
Point 3. Image 10 doesn’t show the positions, distances and overlaps indicated in the realization of case 4.2. The results given in the article differ greatly from those shown in the image. The accuracy of 0.4m is not visible in the image, where much larger errors and great difficulties in the execution of the flight are visible. Can you include an image that allows us to observe compliance with this data?
The reviewer believes that including the case study obtained in the results and the modification of figure 10 should be included in the article.
Response: Figure 10 was constructed separately from localization. As we mentioned in the paper, localization was required to allow the drone avoiding the building during the flight. We added the paragraph where explained this part.
Point 4. The description of acronyms such as RGBD, VSLAM, ROS, RTAB , FAST or FRCNN is missing.
Response: The description is added, thank you
Point 5. In 149 the sSfm algorithm is mentioned, the acronym should be indicated and, if possible, a minimal explanation that allows its inclusion in the article to be understood.
Response: Added, thank you
Point 6. The concept of heterogeneous architecture is used on three occasions, you can explain it on one of the three occasions.
Response: Added, thank you
Point 7. It is necessary to revise, in addition to the English, words that are misspelled such as: perfoem (17), intial (257), recieving (260), please revise.
Response: Done, thank you
Point 8. 5 Tables 2 and 3 should be joined so that the headers can be read and interpreted, in the caution you can indicate the units and make some reference if you cannot include information in the header
Response: Done, thank you
Round 2
Reviewer 2 Report
Thank you for revising the abstract. But it is disappointing that after proofreading there are still many grammar errors. To name a few, on page 2
line 66 "best" article error line 71 "list" article error line 73 "following" article error Please proofread again and pay special attention to missing articles.Author Response
Thanks for pointing on this errors.
We have tried our best to fix all the grammar errors
Reviewer 4 Report
Thanks for your paper remake.
As authors explain, all calculations are made based on a simulation and due to my comments, some parameters can be difficult to evaluate in a simulation. For that reason I would like to know the final process when you pass the research from simulation to a real practice case.
Congratulations for your work.
Author Response
Thanks for your review.
We added in a conclusion the future step - to run on real drones.
