3D LoD2 and LoD3 Modeling of Buildings with Ornamental Towers and Turrets Based on LiDAR Data
Round 1
Reviewer 1 Report
A very good and interesting work
Author Response
Thank you very much for spending your time and efforts reviewing our manuscript.
Reviewer 2 Report
The manuscript is devoted to generating a 3D building model from the LIDAR point cloud. The proposed approach allows automatic building of towers and turrets 3D models. An algorithm exploits the geometry features of the above-mentioned class of 3D models, constructing the rotating surfaces and performing the revolving 3D reconstruction process. I think the proposed approach is computation effective and useful. However, I suggest several comments to improve the readability of the manuscript.
1) Discussion of the proposed approach limitations is very useful.
2) In the proposed approach, five horizontal cross sections have been used. It is necessary to explain such a choice of quantity (why five?).
3) Projection of the point cloud to 2D plane is not a trivial task. That is why the authors should detail this procedure.
4) The authors should detail the possible ways of practice implementation of the proposed algorithm. What program libraries can be used for such implementation?
5) I suggest the algorithm is effective, but the authors should reveal its computation complexity.
6) It is better to explain the approach to visualizations that are shown in Fig.8. What libraries have been used for point cloud and mesh visualizations?
7) What is the output format of the generated 3D models? Could the authors give any supplemental material to demonstrate an example of the 3D model generated using the proposed algorithm?
8) Could the authors give a sample of the source code (progam code) to estimate the effectiveness of the proposed 3D model automatical generation?
I think the comments above can help raise the readability of the manuscript and give readers an opportunity to dive into such an important and hot topic of cost effective 3D reconstruction.
Author Response
The manuscript is devoted to generating a 3D building model from the LIDAR point cloud. The proposed approach allows automatic building of towers and turrets 3D models. An algorithm exploits the geometry features of the above-mentioned class of 3D models, constructing the rotating surfaces and performing the revolving 3D reconstruction process. I think the proposed approach is computation effective and useful. However, I suggest several comments to improve the readability of the manuscript.
Answer:
Thank you very much for spending your time and efforts for reviewing our manuscript. We appreciate all your observations and comments. It is a great pleasure for us to consider all the suggested comments and answering them one by one as follows:
Question 1.1
Discussion of the proposed approach limitations is very useful.
Answer:
A new paragraph is added to Section 6 as follows:
To be fair, it is unavoidable to discuss the limitations of the suggested approach. The proposed algorithm assumes that building facades are completely covered by LiDAR points. In fact, this hypothesis may not be always valid. Therefore, when the façade points disappear for a certain reason, the concerned building details will also disappear. Moreover, the exposed method is very sensitive to noisy points, which can generate a great deformation of the building model. Fortunately, this issue can be solved in future work by enhancing the constructed model though considering the point deviation values in addition to the building symmetry. Some building geometrical forms create hidden areas where laser pulses cannot arrive to certain parts of the building façade such as the case of the building illustrated in Figure 8d. These hidden areas may create distortion in the calculated building model.
Question 1.2
In the proposed approach, five horizontal cross sections have been used. It is necessary to explain such a choice of quantity (why five?).
Answer:
The fifth paragraph in Section 3 is modified as follows:
At this stage, Figure 2b illustrates that the tower is geometrically considered a rotating surface. Regarding the tower body is composed vertically of five main architectural parts (Figure 2), therefore five horizontal cross-sections are calculated from the tower point cloud to verify this hypothesis as shown in Figure 3.
Question 1.3
Projection of the point cloud to 2D plane is not a trivial task. That is why the authors should detail this procedure.
Answer:
The second paragraph in Section 4 is modified as follows:
As the projection will be according to lines parallel to the Z axis, and the projection will be on the horizontal plane OXY, then the result of this operation will be a 2D point cloud where the obtained coordinates will be only X and Y (the same coordinates X and Y in 3D space). In other words, this operation could be realized by considering only the coordinates X and Y to define the new 2D point cloud that represents the tower footprint. Indeed, the mere removing of the Z coordinate from the original point cloud allows obtaining the target 2D footprint point cloud. Figure 5a visualizes the 2D tower footprint point cloud obtained through this operation. It could be observed that due to the irregular distribution of the 3D points on the tower surfaces, the point density of the obtained cloud is irregular too where the right-side density is greater than the left-side one.
Question 1.4
The authors should detail the possible ways of practice implementation of the proposed algorithm. What program libraries can be used for such implementation?
Answer:
The last paragraph in Section 4 is modified as follows:
At this stage, it is important to note that MATLAB software is used to implement the suggested algorithm, then the command “surf (X, Y, Z)” is employed to visualize the calculated building model. Once the modelling suggested approach is detailed, the used datasets will be presented in the next section and Section 6 will present other results of the suggested approach in addition to modelling accuracy estimation.
Question 1.5
I suggest the algorithm is effective, but the authors should reveal its computation complexity.
Answer:
Section 4 is modified as follows:
At this stage, it is important to note that MATLAB software is used to implement the suggested algorithm, then the command “surf (X, Y, Z)” is employed to visualize the calculated building model. However, the pseudocode of the suggested algorithm is as follows:
Input (point cloud (X, Y, Z), m, θ)
ascending sort of “point cloud” according to Z values
i = find(X> Xg -Td and X< Xg +Td and Y ≤ Yg)
SCS = [Y(i), Z(i)]
for i = 1 To length (SCS) Step = 1
for j = 0 To m Step = 1
Zb (i, j+1) = SCS (i, 2)
Xb (i, j+1) =Xg + (Yg - SCS (i, 1)) x cos ( )
Yb (i, j+1) =Xg + (Yg - SCS (i, 1)) x sin ( )
Next j
Next i
Surf (X, Y, Z)
Where SCS is the Semi Cross-Section point list, Surf is the 3D visualization function, Xb, Yb and Zb are the three matrices of the building model (Equations 4 and 5).
In the previous pseudocode, it can be noticed that the suggested algorithm is very short and simple. The algorithm output is three matrices which can be exported in any desired format. Once the modelling suggested approach is detailed, the used datasets will be presented in the next section and Section 6 will present other results of the suggested approach in addition to modelling accuracy estimation.
Question 1.6
It is better to explain the approach to visualizations that are shown in Fig.8. What libraries have been used for point cloud and mesh visualizations?
Answer:
Section 4 is modified as follows:
The last paragraph in Section 4 is modified as follows:
At this stage, it is important to note that MATLAB software is used to implement the suggested algorithm, then the command “surf (X, Y, Z)” is employed to visualize the calculated building model. Once the modelling suggested approach is detailed, the used datasets will be presented in the next section and Section 6 will present other results of the suggested approach in addition to modelling accuracy estimation.
Question 1.7
What is the output format of the generated 3D models? Could the authors give any supplemental material to demonstrate an example of the 3D model generated using the proposed algorithm?
Answer:
Section 4 is modified as follows:
In the previous pseudocode, it can be noticed that the suggested algorithm is very short and simple. The algorithm output is three matrices which can be exported in any desired format. Once the modelling suggested approach is detailed, the used datasets will be presented in the next section and Section 6 will present other results of the suggested approach in addition to modelling accuracy estimation.
Question 1.8
Could the authors give a sample of the source code (progam code) to estimate the effectiveness of the proposed 3D model automatical generation?
Answer:
Section 4 is modified as follows:
At this stage, it is important to note that MATLAB software is used to implement the suggested algorithm, then the command “surf (X, Y, Z)” is employed to visualize the calculated building model. However, the pseudocode of the suggested algorithm is as follows:
Input (point cloud (X, Y, Z), m, θ)
ascending sort of “point cloud” according to Z values
i = find(X> Xg -Td and X< Xg +Td and Y ≤ Yg)
SCS = [Y(i), Z(i)]
for i = 1 To length (SCS) Step = 1
for j = 0 To m Step = 1
Zb (i, j+1) = SCS (i, 2)
Xb (i, j+1) =Xg + (Yg - SCS (i, 1)) x cos ( )
Yb (i, j+1) =Xg + (Yg - SCS (i, 1)) x sin ( )
Next j
Next i
Surf (X, Y, Z)
Where SCS is the Semi Cross-Section point list, Surf is the 3D visualization function, Xb, Yb and Zb are the three matrices of the building model (Equations 4 and 5).
In the previous pseudocode, it can be noticed that the suggested algorithm is very short and simple. The algorithm output is three matrices which can be exported in any desired format. Once the modelling suggested approach is detailed, the used datasets will be presented in the next section and Section 6 will present other results of the suggested approach in addition to modelling accuracy estimation.
Author Response File: 
Author Response.pdf
Reviewer 3 Report
the paper presents an interesting topic but needs some reviews:
1. Lines 45-49 "Classified and segmented point clouds from aerial images fulfil the requirements for building modelling at LOD0, LOD1 and LOD2 levels of detail. Additional data (for example terrestrial scans) are needed to generate LOD3 models, whereas LOD4 modelling relies on indoor scans. Building Information Modeling (BIM) datasets generated in the process of designing, modelling, and managing buildings pose an alternative to the above solutions [14, 17]." needs improvement: why they fulfil the requirements? why BIM is a solution? it is ok maybe is known, but it is always better to shortly explain why.
2. Lines 77-92. Same thing: it is never a good thing to describe a process or methodology without explaining things a little critically.
3. lines 109-110: have previously discussed, not were
4. lines 148-149 why is important to note the use of specific sw?
5. lines 183 - 187 the conclusions are not that clear because the previous lines are not explained well. try to deepen the contents and the topics
6. lines 209 and on: why only 2D information are fine? if you have 3D data why you simplify them?
7. Lines 268-274. The conclusion here are not that simple and not that clear as the authors state. It needs a more accurate critical explanation of the process.
8. How did you do the comparison among the different models? how did you get the results shown in the histogram and table 1?
9. Conclusions need improvements and a more critical approach, they are too simple
Author Response
The paper presents an interesting topic but needs some reviews
Answer:
Thank you very much for taking the time to review our manuscript. We note from your feedback that you reviewed the paper very carefully line by line. We appreciate the great effort you put into this task. We feel extremely fortunate that our paper has received such attention from you. In fact, such high-quality reviewing reflects the outstanding quality of the Remote Sensing Journal. This is why we thank the journal editor for the careful selection of the referee team. At this stage, we would like to confirm that the paper has been revised and edited by considering all your comments. All corrections are now highlighted in yellow. I hope that the revised version meets the required standard and is accepted for publication in this journal.
Question 2.1
Lines 45-49 "Classified and segmented point clouds from aerial images fulfil the requirements for building modelling at LOD0, LOD1 and LOD2 levels of detail. Additional data (for example terrestrial scans) are needed to generate LOD3 models, whereas LOD4 modelling relies on indoor scans. Building Information Modeling (BIM) datasets generated in the process of designing, modelling, and managing buildings pose an alternative to the above solutions [14, 17]." needs improvement: why they fulfil the requirements? why BIM is a solution? it is ok maybe is known, but it is always better to shortly explain why.
Answer:
Thank you for this remark. We have corrected, completed the text. We should write: Classified and segmented point clouds based on aerial photographs meet the requirements for modeling buildings, at the detail levels LOD0, LOD1 and LOD2, as long as their visibility is not disturbed by natural and artificial curtains. Additional data (for example, ground scans) is needed to generate LOD3 models because the details of the extended facades of buildings from aerial scanning are often not well recorded. LOD4 modelling is additionally associated with the modelling of building interiors, and this is based on internal scans.
We also added an explanation related to BIM, writing: Currently, biM technologies are increasingly used in the process of building design and management to create virtual 3D models of buildings with architectural detail. The BIM dataset generated in the process of designing, modeling and building management meets the requirements for creating LOD models at different levels of detail. They are an alternative to the above solutions [14, 17].
Question 2.2
Lines 77-92. Same thing: it is never a good thing to describe a process or methodology without explaining things a little critically.
Answer:
We agree with the remark and this paragraph at the end a few words have been added explaining the essence of the presented modeling methods. "All the method proposals, algorithms, building modeling, at different levels of detail, always give simplified results. Models rarely fully correspond to reality, because construction technologies allow to create complex spatial structures that are difficult to render in 2D or even 3D mathematical recording, used for virtual visualization of entire cities. This forces us to look for new solutions, which is the subject of the research undertaken.
Question 2.3
Lines 109-110: have previously discussed, not were
Answer:
Thank you for this remark. We have changed to: “Huang et al. [35] notice this problem.”
Question 2.4
Lines 148-149 why is important to note the use of specific sw?
Answer:
Thank you for another detailed remark. This is not important, this is our mistake. It was supposed to be information indicating that the authors, when building the model presented in Fig. 2 c, used an official model from Polish GIS imported into CAD, visible in the drawing in the form of a skeleton. Towers in the form of a solid, they drew based on a cloud of points using tools from CAD. They didn't do this model automatically. Correction made were done to the text by changing the sentence.
Question 2.5
Lines 183 - 187 the conclusions are not that clear because the previous lines are not explained well. try to deepen the contents and the topics
Answer:
The fourth paragraph in Section 3 is modified as follows:
Conclusion: Model 1 is more faithful to the tower point cloud than Model 2 because it is constructed using only the point cloud as input. Thus, an automatic modelling approach based only on the airborne LiDAR point cloud will apply the processing procedures and measurements directly on the point cloud, is expected to be more faithful to the point cloud regarding the original building. Consequently, the constructed building may be similar to Model 1 more than Model 2.
Question 2.6
Lines 209 and on: why only 2D information are fine? if you have 3D data why you simplify them?
Answer:
The first and second paragraphs in Section 4 are modified as follows:
The input of the suggested modelling approach is the tower 3D airborne LiDAR point cloud. The presented algorithm aims to construct a 3D model of the scanned tower automatically. Figure 4 summarizes the proposed method which consists of five consecutive steps. First, to calculate the tower footprint gravity center, the tower point cloud is projected on a horizontal plane passing through the coordinate origin.
As the projection will be according to lines parallel to the Z axis, and the projection will be on the horizontal plane OXY, then the result of this operation will be a 2D point cloud where the obtained coordinates will be only X and Y (the same coordinates X and Y in 3D space). In other words, this operation could be realized by considering only the coordinates X and Y to define the new 2D point cloud that represents the tower footprint. Indeed, the mere removing of the Z coordinate from the original point cloud allows obtaining the target 2D footprint point cloud. Figure 5a visualizes the 2D tower footprint point cloud obtained through this operation. It could be observed that due to the irregular distribution of the 3D points on the tower surfaces, the point density of the obtained cloud is irregular too where the right-side density is greater than the left-side one.
Question 2.7
Lines 268-274. The conclusion here are not that simple and not that clear as the author’s state. It needs a more accurate critical explanation of the process.
Answer:
Section 4 is modified as follows:
Figure 6 illustrates the 3D model of the tower point cloud shown in Figure 2a. This model is constructed automatically using the approach described above which is based on the rotating surface. It can be noted that this model is similar to the model shown in Figure 2b because the two models consider only the tower point cloud. Indeed, the two models represent a rotating surface and consist of five parts vertically superimposed. Furthermore, their dimensions are similar to the mean dimensions measure directly from the point cloud. In contrast, the model shown in Figure 2c which considers the tower point cloud in addition to the tower image has a great difference regarding the calculated model.
Question 2.8
How did you do the comparison among the different models? how did you get the results shown in the histogram and table 1?
Answer:
A new paragraph is added to Section 6 as follows:
At this stage, it is important to underline that the values of Table 1 are calculated in the same way as the calculation of the histogram illustrated in Figure 7. Hence, the tower point cloud is superimposed on the calculated model, and then the deviation of each point regarding the constructed model is calculated. This operation allows obtaining the deviation value for each point. The calculation of histogram of all deviation values produces Figure 7 whereas the analysis of deviation values allows calculation of Table1.
Question 2.9
Conclusions need improvements and a more critical approach, they are too simple
Answer:
Section 7 is modified as follows:
This paper proposed an automatic modelling approach based only on the airborne LiDAR point cloud and applied the processing procedures and measurements directly on the point cloud. Thus, the constructed models were more faithful to the point cloud regarding the original building. The suggested approach focused only on the buildings having rotating surface geometric forms. To construct the building model, it assumed that the building is symmetric around the rotating axis, which is why it calculated the building's vertical cross-section before rotating it to generate the 3D building model. The modelling algorithm used the matrixial form to present mathematically the constructed building model.
To evaluate the accuracy of the suggested method, five-tower point clouds were employed. Hence, the deviations of points regarding the obtained model were calculated in addition to the standard deviation factor. Despite the efficacity of the suggested algorithm, it suffered from three main limitations. In fact, it was supposed that the tower is entirely covered by LiDAR point which is not always valid. Also, some tower geometry forms may generate hidden areas which can produce undesirable deformation in the tower model. Moreover, the suggested algorithm is sensitive to the presence of noisy points. However, the façade decorations, the noise presence, the LiDAR point accuracy, and the noisy points played an essential role in decreasing the constructed model’s accuracy. The building model can be enhanced in future work by considering the point having considerable deviations. Indeed, the used matrixial form enables to carrying out of local enhancements. Moreover, the use of vertical cross-section in addition to the matrixial form will be a research topic to develop a new building modelling approach regardless of the architectural complexity level of the building. Finally, additional data may be integrated into the envisaged modelling approach such as areal as well as terrestrial imagery to make it more faithful to the original building.
Author Response File: Author Response.pdf
