Digital Outcrop Model Generation from Hybrid UAV and Panoramic Imaging Systems
, Ademir Marques Junior 1, Daniel Capella Zanotta 1, André Luiz Durante Spigolon 2, Mauricio Roberto Veronez 1 and Luiz Gonzaga, Jr. 1
Round 1
Reviewer 1 Report
The manuscript “Digital Outcrop Model generation from hybrid UAV and panoramic imaging systems” presents a successful attempt to integrate framework between terrestrial Spherical Panoramic Images (SPI), acquired by omnidirectional fusion camera, and UAV survey. In this way the authors overcome gaps left by traditional mapping in complex natural structures like outcrops. The novel method for adaptive integration is made through an optimized selective strategy based on an octree framework. The authors present experimental data to verify the method suggested. The assessment of the model is done by quantitative and qualitative indicators. The results show the potential of generating a more reliable 3D model using SPI allied with UAV image data while reducing field survey time and complexity. Further improvements of the proposed method could be done.
The manuscript needs one obligatory correction: The caption of Figure 9 should be corrected. The notation of b), c) should be corrected, while d), e) and f) should be explained.
Author Response
The authors would like to thank the reviewer for his/her valuable criticism and time spent in reviewing our work. Please, find in what follows detailed replies regarding questions raised during the revision. To facilitate the 2nd round of the revision, corresponding changes in the current version of the manuscript are highlighted in red.
The manuscript “Digital Outcrop Model generation from hybrid UAV and panoramic imaging systems” presents a successful attempt to integrate framework between terrestrial Spherical Panoramic Images (SPI), acquired by omnidirectional fusion camera, and UAV survey. In this way the authors overcome gaps left by traditional mapping in complex natural structures like outcrops. The novel method for adaptive integration is made through an optimized selective strategy based on an octree framework. The authors present experimental data to verify the method suggested. The assessment of the model is done by quantitative and qualitative indicators. The results show the potential of generating a more reliable 3D model using SPI allied with UAV image data while reducing field survey time and complexity. Further improvements of the proposed method could be made.
The manuscript needs one obligatory correction: The caption of Figure 9 should be corrected. The notation of b), c) should be corrected, while d), e) and f) should be explained.
Author’s response: We appreciate the accurate summary regarding our work. Thanks also for drawing our attention to the mistakes committed in Figure 9. We fixed the notation of the figure accordingly. Please, find changes in red.
Reviewer 2 Report
Review paper for “Digital Outcrop Model generation from hybrid UAV and panoramic imaging systems”
The paper presents the method integration between Spherical Panoramic Images (SPI) and UAV survey to overcome gaps left by traditional mapping in complex natural structures like outcrops. These data are extracted from real capture. All results showed the performance of the paper proposal. This topic has some interest in image processing as well as in mapping fields. In general, the paper content is within the scope of the journal. However, there are some points that need to be addressed in detail to meet the quality of the publication. The comments and suggestions are given below:
1. Eq. 5 must be considered carefully. It presents for standard deviation equation.
2. The toolbox for reducing 20% of points is 7.154.368 points. Your calculation in line 273 needs to be re-write in the revised manuscript.
3. Your results used all points (point cloud) in offline mode. Please discuss in an online case the speed of process and memory for storing all point clouds. Although the author mentioned in line 315 of memory (offline), it should be explained in the detail.
4. Fig.9 showed the performance of SPI. However, the author must explain what is improved in detail by some plots (quantitative).
5. Please organize the paper again and present more contributions in the revised version instance describing all common information.
Finally, a major revision is needed to rewrite the paper to reflect the contribution properly before the paper might be accepted for publication the above issues should be addressed.
Author Response
The authors would like to thank the reviewer for his/her valuable criticism and time spent in reviewing our work. Please, find in what follows detailed replies regarding questions raised during the revision. To facilitate the 2nd round of the revision, corresponding changes in the current version of the manuscript are highlighted in red.
Reviewer 2
Review paper for “Digital Outcrop Model generation from hybrid UAV and panoramic imaging systems”
The paper presents the method integration between Spherical Panoramic Images (SPI) and UAV survey to overcome gaps left by traditional mapping in complex natural structures like outcrops. These data are extracted from real capture. All results showed the performance of the paper proposal. This topic has some interest in image processing as well as in mapping fields. In general, the paper content is within the scope of the journal. However, there are some points that need to be addressed in detail to meet the quality of the publication. The comments and suggestions are given below:
- Eq. 5 must be considered carefully. It presents for standard deviation equation.
Author’s response: Thanks for drawing our attention for this unnoticed mistake. The standard deviation equation is now corrected.
- The toolbox for reducing 20% of points is 7.154.368 points. Your calculation in line 273 needs to be re-write in the revised manuscript.
Author’s response: The reviewer is indeed right. We have forgot to mention an intermediary result between the original dense cloud and the cleaned one, which is the rough delimitation of the study area. The current version of the manuscript is now reflecting this step.
- Your results used all points (point cloud) in offline mode. Please discuss in an online case the speed of process and memory for storing all point clouds. Although the author mentioned in line 315 of memory (offline), it should be explained in the detail.
Author’s response: The previous version was a bit incomplete regarding computational costs. We now give further examples of storage (off line) and memory usage (on line). To quantitatively measure the optimization reached with our technique, we now report the execution time when running the kd-tree algorithm in both original and optimized dense clouds.
Please, find it in red in lines 324-328.
- Fig.9 showed the performance of SPI. However, the author must explain what is improved in detail by some plots (quantitative).
Author’s response: To comply with the reviewer’s suggestion, we now prepared a new figure (Figure 10) reflecting the exact places where the number of points in the cloud were increased. Clarification about the process can be found in line 267.
- Please organize the paper again and present more contributions in the revised version instance describing all common information.
Author’s response: As an attempt to tackle your suggestion, we have summarized our major contributions in the discussion section of the paper, which were not as clear as it should be in the first submitted version of the manuscript. As the reviewer may find in the added paragraphs, the paper stands out in the topic of photogrammetry for DOM generation for using an omnidirectional camera, which is a novelty in the geology field, even though other areas have already been studying this approach recently, as mentioned in related works. Also, the developed script for dense cloud merging is now made available for free at Github for anyone to use and contribute to the code. The link to the repository was added as a reference in the paper in line 266.
Finally, a major revision is needed to rewrite the paper to reflect the contribution properly before the paper might be accepted for publication the above issues should be addressed.
Author’s response: We thank the reviewer for his/her valuable suggestions. We changed many parts of the paper in order to clarify key points of the contribution in the revised version.
Reviewer 3 Report
[general comments]:
The paper presents a valid contribution in selecting different sources for the points used in the reconstruction.
Comparing the final geometric quality with a TLS survey would add to the quality of the paper.
<please add comments regarding TLS/ground truth and relative quality assessment used>
The paper mentions 280 UAV photos, although the genus of the object is approximately zero, maybe the quality of the model could be improved if a few thousand images were acquired instead (perhaps in future work, an experiment with a similar object and 1000 vs 280 photos could be used to compare).
The rough dimensions of the object appear to be missing, in Fig 4c, one can guess that it is less than 50x50x50 meters, but if you have that information, it helps to contextualize the acquisition, planning.
line 276-281
holes in UAV data often produce larger triangles when they are meshed, perhaps this property could also be tested in the future
[minor comment]:
line 79, [15] in addition to 15, could also cite:
Covas, J.; Ferreira, V.; Mateus, L. 3D reconstruction with fisheye images strategies to survey complex heritage buildings. In
2015 Digital Heritage 1, 123-126.
which is particularly useful for the image acquisition in tight stairwells.
line 188 "(SOR) tool of the software is applied"
to make it more clear, instead of software, write the name of the software again
line 241
"The octree" there is no citation, consider citing perhaps the easiest implementation of an octree, where vertices
are sorted in the array in-place according to their 1-8 volume-id:
Oliveira, J.; Buxton, B. An efficient octree for interactive large model visualization. RN-05-13, 2005.
http://www0.cs.ucl.ac.uk/research/researchnotes/documents/RN_05_13.pdf
line 325
downwards (blue) and upwards (red)
caption of Figure 9
(b) should be (C)
and
(c) should be (e)
Author Response
The authors would like to thank the reviewer for his/her valuable criticism and time spent in reviewing our work. Please, find in what follows detailed replies regarding questions raised during the revision. To facilitate the 2nd round of the revision, corresponding changes in the current version of the manuscript are highlighted in red.
Reviewer 3
The paper presents a valid contribution in selecting different sources for the points used in the reconstruction.
Comparing the final geometric quality with a TLS survey would add to the quality of the paper.
<please add comments regarding TLS/ground truth and relative quality assessment used>
Author’s response: We appreciate the interesting reviewer suggestion regarding using TLS data in the validation of the results. We understand TLS is being used in parallel with multi-view photogrammetry to produce digital elevation models. In fact, some recent studies have addressed accuracy comparison between digital elevation models from both sources. Thus, we decided to focus on photogrammetric products only, so avoiding more complexity to the problem at this moment (comparison among three different sources – TLS, SPI, UAV photogrammetry). Although, we consider using TLS (which also is expected to present issues covering subvertical features) in future lines of research.
In the present version of the manuscript, we added some phrases to clarify the reason to not consider TLS for validation at this moment.
The paper mentions 280 UAV photos, although the genus of the object is approximately zero, maybe the quality of the model could be improved if a few thousand images were acquired instead (perhaps in future work, an experiment with a similar object and 1000 vs 280 photos could be used to compare).
Author’s response: We once more appreciate the reviewer’s suggestion. One of our research lines address optimization of the reconstruction by controlling number of points in different location/targest based on its relevance to the application (in our case, outcrops). Decimation of the mesh and selective filtering of the points play a crucial role on final quality of the models. In our lab we use to load the reconstructed models in an immersive VR environment, and thus performance of the system is dramatically impacted by excessive amount of points (which conversely improve quality). In fact, there is a complex trade-off between using many images (number of reconstructed points) and a suitable model to be managed by the system.
The rough dimensions of the object appear to be missing, in Fig 4c, one can guess that it is less than 50x50x50 meters, but if you have that information, it helps to contextualize the acquisition, planning.
Author’s response: Thanks for drawing our attention to this important missing information. Indeed, size of the object and number of photo acquisitions should be informed clearer to allow a correct comprehension of the problem. In the current version this information is added (red).
line 276-281
holes in UAV data often produce larger triangles when they are meshed, perhaps this property could also be tested in the future
Author’s response: This property is very important and is somehow present in the analysis of our study in figure x. The comparison using C2C converts one of the clouds to mesh and interpolate the areas with less points creating larger triangles. These areas are showed in color red and helped us to notice where the original point cloud provided poor results that should be improved by applying SPI data.
[minor comment]:
line 79, [15] in addition to 15, could also cite:
Covas, J.; Ferreira, V.; Mateus, L. 3D reconstruction with fisheye images strategies to survey complex heritage buildings. In
2015 Digital Heritage 1, 123-126.
which is particularly useful for the image acquisition in tight stairwells.
Author’s response: Thanks for the suggestion. This reference was added later in the line 144.
line 188 "(SOR) tool of the software is applied"
to make it more clear, instead of software, write the name of the software again
Author’s response: We now make it clearer citing the software name.
line 241
"The octree" there is no citation, consider citing perhaps the easiest implementation of an octree, where vertices
are sorted in the array in-place according to their 1-8 volume-id:
Oliveira, J.; Buxton, B. An efficient octree for interactive large model visualization. RN-05-13, 2005.
http://www0.cs.ucl.ac.uk/research/researchnotes/documents/RN_05_13.pdf
Author’s response: We added the reference to the line 242.
line 325
downwards (blue) and upwards (red) caption of Figure 9 (b) should be (C) and (c) should be (e).
Author’s response: The Figure and text is now corrected on the line 338.
Round 2
Reviewer 2 Report
The revised manuscript already solved all questions of the reviewer. Please consider accepting this manuscript.
