Semi-Automated Segmentation of Geometric Shapes from Point Clouds
Round 1
Reviewer 1 Report (Previous Reviewer 2)
Thanks for your efforts and wish you all the best.
Author Response
Dear reviewer,
thank you very much for reviewing our paper. We really appreciate it.
Reviewer 2 Report (New Reviewer)
Dear authors,
you have written very interesting article according to my opinion, and to my field of interest. English language is fine and doesn’t need any improvements. You cited all the literature correctly. I didn’t find any major flaws in the article, i.e., in the methodology, presentation and results.
The good points of the article are:
1. very interesting topic regarding laser scanning and segmentation of geometric shapes from point clouds
2. very detailly elaborated approach for point cloud segmentation
According to my opinion, the weak points of your article are:
1. design of the article from section 3
2. conclusion is too short
The improvement should be done in the following:
Ad. 1) Design of the article from section 3. should be rearranged as follows, according to my opinion.
Section 3. belongs to the methodology and should become section 2.4.
Section 4. Testing of the proposed approach should be divided into two sections:
- section 3. Testing of the proposed approach. Also, here you should elaborate the point cloud no. 3 because it is not elaborated here, and from line 581 it is used for the analysis and comparison of proposed algorithm versus RANSAC algorithm.
- section 4. Results and discussion – here you should move the section 4.1 with its subsections, now 4.1. Plane segmentation, 4.2. Sphere segmentation and 4.3. Cylinder segmentation.
- In this new section 4. you should also add discussion and discuss achieved results. Here it could also be interesting to compare your proposed algorithm with other algorithms in general - use references no. 7, 17 and 22.
Ad. 2)
The conclusion is too general, and actually it more sounds like abstract of the article. You should rearrange it. Please expand the conclusion with the main advantages of your algorithm based on the performed tests. Briefly enumerate performed tests and summarize achieved results.
Suggestion by “lines”:
- all text: please use the symbol s for standard deviation after first mentioning in all tables and text.
- line 290-291: please enumerate these several conditions for stopping the calculation.
- line 411 – which scanner was used, please specify. It is known from line 442, but please specify here also. Further, what was the average point cloud density, and the accuracy in the spatial position of a single measured point?
- line 420: the differences among the known parameters – how these known parameters where obtained, i.e., from where are they known?
- line 469: the known (real) parameters of individual cylinders – how these known parameters where obtained, i.e., from where are they known?
- line 472-473: The parameters of each of the columns were measured at various positions – how where they measured and what was the accuracy of used instrument?
- line 505: known geometric parameters - how these known parameters where obtained, i.e., from where are they known? You specified this in lines 507-509. Please specify also for the fist two tests, as I mentioned above.
- line 509-511: it would be better if you elaborated the results in the table as you did for the first and second test.
- line 541: please expand the heading of the figure – left and right
- line 570: please expand the heading of the figure – left and right
- line 582: Point Cloud No. 3 – it is mentioned for the first time. Why wasn’t it elaborated in section 4? In section 3 you elaborated point clouds 1, 2 and 4. Point cloud 3 should also be elaborated there.
Best regards.
Author Response
Dear reviewer,
The authors would like to thank you for providing valuable feedback that we believe will greatly improve the quality of the paper. Please find our responses to your comments in attached file.
Best regards.
Author Response File: Author Response.pdf
Reviewer 3 Report (New Reviewer)
As far as I know, this is a novel adaptation of existing RANSAC/region growing approaches for segmenting point datasets. The method is clearly described and the experimentation is convincing.
However, a couple of aspects of the proposed method and its experimentation could have been improved. The manuscript is riddled with taylor-made parameters:
-LNV < 1º or plane detection, < 5º for sphere detection or < 3º for cylinder detection.
-"... 50% of the points need a higher LPD than the ideal LPD".
-"If 100 incorrect attempts ... are made in a row ... the calculation is stopped".
-"The iterative re-estimation is performed until all the points of the detected cylinder have been selected, and the maximum number o iterations is set to 15".
-"If at least 25% of the cells have ideal coverage, the cylinder is considered a reliable one"
etc.
This erodes the robustness of the method and narrows its scope of application. A further study proposing rules to estimate some of these parameters for any given dataset (instead of providing values estimated empirically) would have lead to a much more sound method.
My second concern is about the comparison with a standard RANSAC method in Section 4. It seems the authors used an in-house implementation. Comparing with the available implementation of the optimized RANSAC by Schnabel et al. [Schnabel2007] would be a much better validation for the proposed approach. By the way, this reference should be included in Section 1 since this is probably the state-of-the-art in RANSAC-based approaches. And also this is a VERY GOOD EXAMPLE of a paper that provides methods for computing the critical parameters of the algorithm instead of actual values.
Anyway, any of these two improvements would require a significant amount of work, therefore, I suggest accepting the paper as it is, although I believe the authors have wasted an opportunity to give much more exposure to their work.
REFERENCES
[Schnabel2007] Ruwen Schnabel, Roland Wahl, and Reinhard Klein "Efficient RANSAC for Point-Cloud Shape Detection" In: Computer Graphics Forum (June 2007), 26:2(214-226)
Author Response
Dear reviewer,
The authors would like to thank you for providing valuable feedback. We agree with your comments, and in the future, we will improve the approach because it will be a part of an approach for as-built documentation of buildings using BIM and TLS data.
Regarding the paper Schnabel et al., it is cited in section 1 also in the original version of the paper, and it is included in the list of references as [21].
We believe that your comments will improve our future work.
Round 2
Reviewer 2 Report (New Reviewer)
Dear authors,
thank you for accepting my suggestions.
Best regards.
Author Response
Dear reviewer.
Thank you once again.
Best regards.
This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.
Round 1
Reviewer 1 Report
Automated Detection of Geometric Primitives in Point Clouds
This is a revision of a resubmitted paper.
The authors have addressed all of my comments from previous revisions, therefore I suggest accepting the paper.
The paper should be proofread. Especially the newly added parts of the manuscript could use a rewrite by a more experienced English speaker.
Minor comments:
- sometimes you use Fig., sometimes Figure ... be consistent.
- line 563 ... there is no "." after Figure number.
- line 565 ... "Figure. 14" a typo.
- line 568 ... "Table 2." ... obsolete "." after Table number.
- line 574, 579 ... "Figure 16." ... obsolete "." after Figure number.
- line 603-604, 611 ... "Figure 10., and Figure 17." ... there are obsolete "." after Figure numbers.
Reviewer 2 Report
Dear Authors
thanks for your efforts in producing this nice work.
it would be great I’d you can consider my comments to enhance the quality of your product.
thanks and best regards
Comments for author File: Comments.pdf
Reviewer 3 Report
The article deals with the well-known topic of automated segmentation of geometric shapes from point clouds. Although it has been described many times in the literature, the improvement of these methods is important in the context of the development of BIM models and their increasingly wider use in practice. The limitations of the presented algorithm are the necessity to pre-select the searched geometric primitives and the need to adopt certain parameters (thresholds). Some parameters are adopted empirically. The advantages of the article are a fairly detailed description of the proposed algorithm and in the algorithm there is no need to select exactly the number of geometric shapes in the point cloud.
A detailed list of proposed amendments or comments to the manuscript content is provided below.
The cited bibliography is mostly outdated, mainly from the years 2001-2013. It is necessary to update the literature review by adding newer items.
The title of the article should contain the word "semi-automated" as the Authors themselves indicate when describing their proposed algorithm (e.g. lines 659-660)
The Introduction has a rather unusual layout. In lines 51-80, there is a very cursory description of the segmentation methods (divided into five categories). Then these methods are described in the following paragraphs. I encourage the Authors to consider merging these two parts of the text. In one paragraph should be a complete description of a specific category of segmentation methods.
What do the parts of manuscript marked in yellow mean?
Lines 21-25. There is no comparison to the results from RANSAC in the abstract
Lines 16-18. Please check the structure of the sentence
Lines 44-47. The sentence is imprecise. There is no explanation as to what the authors mean by suggesting the use of detected shapes in TLS? Please specify it.
Lines 63-68. Attribute-based methods are very briefly described. This description should be supplemented.
Line 89. Mistake in reference to literature; should be [25]
Line 152. k-nearest... "k" should be in italics
Line154 and Equation 1. Inconsistent formatting of dotNorm: italic or simple font, subscript
Line339 and Equation 7. Inconsistent formatting of ctop and cbottom
Equation 8. "&" is not a mathematical symbol
Lines 346-347 and Equation 8. Inconsistent formatting of td i tn
Line 513 The numbering of point clouds (e.g. Point Cloud no. 1) was only introduced when the results were compared with the RANSAC algorithm. Previously, the same cloud had no numbers, only names (e.g. "point cloud from an industrial building " - lines 398-399). I propose to standardize the marking of point clouds which are common to both chapters (4 and 4.1).
Please standardize units throughout the text of manuscript, e.g. "distance filter was 50 mm" line 406 and "distance threshold = 0.05 m" line 514
Line 546 (first column in Table 2) and Equation 9. Inconsistent formatting of symbols
Line 555. Incorrectly formatted reference (Fig. 7) - should be Figure 7.
If the figure contains several parts, captions (a), (b), ... are usually used instead of the terms "On the left side of the figure" (e.g. line 564)