Model Scaling in Smartphone GNSS-Aided Photogrammetry for Fragmentation Size Distribution Estimation

Round 1

Reviewer 1 Report

The paper describes a method to scale 3D structure from motion data using GNSS data from phones. While the method seems to be interesting for practical application, the presented studies and results seem to be not quite convincing and comprehensive. The reviewer suggests to redesign the studies to account for the nature of inaccurate positional data. More review details are given in the attached document.

Comments for author File: Comments.pdf

Author Response

Greetings,

First and foremost, we would like to thank you for your interest and for taking the time to review our paper. We have tried our best to respond to each and every comment that you have made, and hope that it meets your standards. Please refer to the attachment for our point-by-point response.

Author Response File: Author Response.pdf

Reviewer 2 Report

General Comments

This is a really interesting paper evaluating accuracy of SfM/MvS models scaled by low resolution GNSS derived camera positions.  This is one of several very important resolution issues that have arisen in this new technique and as such this paper will get a lot of interest.  That is, there are precious few experiments of this type.  I compliment the authors on giving one of the best descriptions of the quantitative side of SfM and MvS in a compact form—I had to read much of a book to get comparable information that can be gleened from their nice figures.   So bravo there.

Although I really like the essence of the paper, I think a few revisions would really increase the impact of this paper.  That is, some minor tweeks could make this into a landmark paper on the issues raised.  I see a few key things needed in this paper:

• I felt like a reader of a whodunnit novel getting to the end and find that the answer to the main question remained unclear—like a trial that ends with an undecided jury. What the paper desperately needs is a discussion of the results of the experiments and why the last experiment (with the pseudorock pile) went abruptly from having a pretty lousy solution to an impressive solution by simply doubling the number of photos.  The other experiments suggested that linearly increasing the number of cameras would linearly increase  resolution, but in that experiment it was clearly nonlinear—any idea why?  That is what should be discussed, rather than just making and observation in the conclusions.
• I think there should be more of a discussion up front about scale of the models being developed vs GNSS error. https://doi.org/10.1016/j.jsg.2018.05.014 ran into this problem head on when they tried to image an outcrop ~100m across with GNSS location. Their main result was that they had large rigid body rotation errors due to their imaging array (a serendipitous experiment because they didn’t think of the problem ahead of time, and not that relevant here), but they also realized they had scaling issues and geometric warping related to GNSS errors.  That led them to the “oh, they were not thinking this through” realization that GNSS error was a contributing factor because the GNSS error was a significant fraction of the scale of the model.  That is almost certainly at play here in the pseudomuck pile experiment—it is irrelevant to the lab experiment where positions were very accurately recorded by the grid.  They have found (work in prep) that problems like that disappear at large scale, presumably because the gnss error gets small relative to the model size.  I wonder if this gnss error vs model size is what is happening here in the pseudomuck pile model?  I think it is likely a factor and needs to be discussed.
• A small issue is a couple bookkeeping things. One of them is you never state what software you are using.  For those of us who are more users than developers, that is an important question.  For example, we use both pix4D and metashape, and although they usually yield broadly similar models, there are distinct differences.  No need to rerun your experiments with different software, just state what you’re using.
• A somewhat different issue is there is an alternative solution to the main problem posed in this paper. That in no way negates the importance of this paper, but there are two simple ways to deal with model scaling that we have used.  One is within SfM/MvS software like metashape—placing a scale bar in the scene, or use a known pair of features as a scalebar.  That allows automatic scaling within the software.  A related solution (which also requires a scale bar) is after processing the data, you can use a scale bar to rescale the model in other software, like cloud compare, or maybe more important for mining applications, maptek’s point studio.  That is a simple procedure, albeit requiring an additional step, but you should at least admit this alternative.
• Finally, another trivial point, but distracting, is a style issue. The fact I can even say this is a complement to non-native English speaking authors, but use of the future tense in scientific writing can lead to confusion.  When you use future tense it isn’t clear you are referring something you are going to do in the future (not part of this paper) vs something within this text.  So it is better to use past and/or present for the text to avoid this confusion.   

In summary, I really like this paper despite the points I make above.  I think these fixes are easy and if the authors add a few sentences on some of these points it will really increase the impact of the paper, raising it to the level it deserves. 

Specific comments keyed to lines in the text:

57:  Might make a brief statement here about the size of the muckpiles being analyzed.  Beyond the specifics reported here, those of us involved in other SfM related experiments would benefit greatly from knowing what the range of scales are that you’re analyzing.  For example, is the muckpile a km across, or is it 10m (I’d assume probably something in between, but would be nice to state)  Also, state clearly here what data type you are analyzing in the analysis—are you using a point cloud, a tin, or both?

66: I don’t see a red box in figure 1—aah, I see; it is Figure 2!!!  Relabel please!

79:  “However” at the beginning of this sentence is unnecessary, or even confusing.

81:  this is a style thing done differently by people, but I am not a fan of using the future tense within a paper.  To the reader this is “now”, so the present tense is more compact and easily understood, particularly by nonnative English speakers.

107:  Maybe I’m being math challenged here but I don’t understand the equation.  I understand the principal where the world coordinate is transformed into the camera coordinates (and the inverse) through a rotation and translation, and scaling.  However, the equation is written like a matrix operation, but it isn’t a proper matrix operation.  I think the reason is there are higher order quantities buried in here (e.g. R and t are stated as vectors, so that 2x2 matrix containing R and t seems like it is probably a third order tensor?).  I think you (authors) are trying to make this compact because most readers eyes will just pass over the equation going to the bottom line.   I wonder if maybe the equation were presented in Einsteinian tensor notation, would that clarify?  I just don’t follow it this way.

205:  positions is a preferable term to “pose” in this context (and an earlier one).  Pose could be misconstrued by readers to mean something different.

224:  The “although” is awkward here.  Although is used to differentiate a dependent clause, not to begin an independent clause in English.  You could just remove the period and replace it with a comma (and remove capital A), but then the sentence is a bit of a run on sentence.  Consider rewriting the two sentences; or merge and don’t worry about the run on sentence.

251-252:  Awkward sentence; not sure what you mean.  Might be mixing tenses—future here, next sentence is past.  I’d try to rewrite but might write it wrong

261:  this result is really interesting, but the error is actually bigger than I would have thought given the carefully controlled experiment.  For most studies and error of 2% is no problem (anything I do, that would be fine with me), but some people use these systems for geodetic work; e.g. surface change studies.  It might be worth adding a sentence here to help direct readers attention to this issue?? (authors call here, this is a mere suggestion to help a broader audience)

280-285:  paragraph suffers from mix of tenses; stick to the present or past tense.  Also, 284—not sure what you mean by 50% weight will be used

307:  It might be worth mentioning that you gnss data were collected beneath partial tree cover (you can see this in the figures), which could be affecting positional accuracy.  Kind of a wild card in the experiment, but worth noting

319-324:  run on sentence hard to follow.  Make it into two sentences?

350:  Wow, the preceding sentences really leave the reader hanging and beg for a discussion of the meaning of your results here!!!

370-end:  but WHY? Why were did the pseudomuck pile experiment show the nonlinear convergence on the right answer?  You really need to discuss what some of the causes are for you observation!!!

Comments for author File: Comments.pdf

Author Response

Greetings,

First and foremost, we would like to thank you for your interest and for taking the time to review our paper. We have tried our best to respond to each and every comment that you have made, and hope that it meets your standards. Please refer to the attachment for our point-by-point response.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Dear authors!

Thank you for submitting the revised manuscript. The reviewer highly appreciates the consideration of some of the points.

Unfortunately, the key concerns about the design and execution of the experiments and field tests have not been eliminated. Please find further information in the attached document.

Comments for author File: Comments.pdf

Author Response

Dear Reviewer,

We would like to thank you for taking the time to review our paper once more and give valuable insights. We hope the revisions that have been made improved the paper to a more publishable standard. As such, the authors welcome any further revisions.

Sincerely,

Zedrick Paul L. Tungol

Doctoral Student

Graduate School of International Resource Sciences

Akita University

Author Response File: Author Response.docx