Regularization of Building Roof Boundaries from Airborne LiDAR Data Using an Iterative CD-Spline
Round 1
Reviewer 1 Report
1. Figure 2 - methodology - dividing point clouds into bulidings and no-buildings is rather segmentation - even classfication than filtering.
In automatic filtration algorithms (only filtration) there is no posiibility to say that this detected structure has to be the bulding. Only additional information (attributes and geometry information) can give the final answear, but this is classification.
This exatly Authors showed in figure 3b - this is only ground and no-ground.
2. line 337 page 9: Authors should at the beginnig say that this is airborne LiDAR data. LiDAR is also terrestriual laser scanning and Differential LiDAR and many others.
the same - page 17 line 491
3. Conclusions: is this method robust and why if not or yes, what is the time of processing according to other existing methods? What is the level of automatization?
My additional question, if using information about the point intensity can be halpful in your approach?
Author Response
Please see the attachment
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The proposed method was interesting to the readers. The paper was well written. The description of the method was clear. However, the designed method has big drawbacks. One is that the boundary extraction method (Adaptive Alpha-shape method) was based on 2D method; another is that the proposed method works well for the curved building edges but not suitable for modelling the rectangular buildings, especially when the building edges have many rectangular turning points. Based on the above facts, the proposed method is only suitable for simple buildings with the curved edges.
Specific:
Page 6, line 218, the authors mentioned that ‘The approximate boundaries were extracted using the adaptive alpha-shape algorithm approach proposed by Santos et al. [30]’. Santos et al. [30] mentioned that the alpha-shape algorithm works for 2D data. The method can only find the outer boundaries. When the buildings are connected to each other and the roofs have the height differences, for example, the B18 in Figure 8 and 12, the internal edges were missing. The 3D representation of B18 was missing in Figure 12. The B18 has rectangular edges shown in Figure 8. From the 2D result in Figure 12, the edges were not rectangular.
Page 2, line 116-122, format issue: adjust the line distances.
Reference needs to be checked. References in the article don’t match with the list. For example, page 6, line 210, Sampath and Shan [9] should be Sampath and Shan [12].
Please check the header and footer information: ‘Remote Sens. 2018, 10, x; doi: FOR PEER REVIEW’
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
This study employed existing mathematical as well as data processing tools in forming a workflow for regularizing building roof boundary from airborne LiDAR point clouds using changeable degree of spline. Building roof boundary points were extracted from LiDAR point clouds, Tdis and Tang were set as thresholds in selecting knot points (critical points) for spline on a segment basis, and an iterative computational algorithm in determining the order of fitting function was implemented and the solutions were found confining to F test.
The proposed workflow is appealing and seemingly workable. The key element in succeeding this work explicitly points to the corner points of building roof in two specific aspects: one is if the corner points have been rendered in point clouds, and the other is how effective and complete the corner points are selected for regularization mission. Revealed from the experiments, most of corner points were identified and supported fair outlines in portraying the building roofs, proving the applicability of the proposed method together with the thresholding parameters under consideration. On the other hand, the failure or unsatisfactory cases indicated the insufficient participation of crucial corner points which, in turn, suggests the need of a better strategy in restoring all crucial corner points as long as they are inherited from point cloud, and that would earn this work a significant uniqueness as compared to the counterparts in the roof reconstruction trend.
The other suggestions toward bettering the current version include the following:
- Serving as a reference data, the metric quality of boundary points by means of photogrammetric approach has not been evaluated, shielding the quantitative comparison. Furthermore, the lack of giving the locations of manually measured points makes the justification of evaluation difficult.
- Building B18 with parapet structure posed relatively poor performance in Table 1 and also was missed (B22 too) from the 3D representation in Fig. 12(b). Besides, no any discussion has been given to this building.
- The significantly large value of RMSEY of B8 in Table 1 may need further explanation.
- English check: Line 229, 238~240, 395~397.
- Last but not least, although the thresholds set showed no significant difference in terms of regularization results both quantitatively and qualitatively, the methodology in choosing optimal parameters taking different data acquisition and types of building roof into account leaves no clear guidance and more effective and sufficient corner points selection awaits actual implementation to realize the unique contribution of this study.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 4 Report
The authors present a manuscript about 3D building modelling based on LiDAR data. The proposed method describes differently shaped buildings defining automatically the polynomial degree that better approximates the boundary. The point boundary of a building is extracted by means of a a-shaped algorithm, while its critical points are detected using an angle based generalization. Segments between two critical points are described through a CD-spline approach and finally an interesting strategy for polynomial degree evaluation is explained.
The paper is well organized: 24 different buildings from simple to more complex shaped ones are presented. Different densities are used; the lowest one is 2.9 pnt/m2. A qualitative and quantitative analysis of the result is presented, moreover the influence of the three used thresholds is verified.
The following points should be better explained before publication:
- A description of the a-shaped algorithm used for point boundary extraction (reference [30])
- A description of the angle regularization algorithm highlighting the differences (if there are) between this method and the one presented by Lee [13].
- Row 165: j starts from 1?
The manuscript can be accepted with these minor revisions.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report
Dear Authors,
Thank you for your efforts and your contributions for our scientific community!
Before publishing this paper, I suggest a change in the title of the paper.
You used 'Regularization of Building Roof Boundaries .....'. It gave the wrong impression that the building roof boundaries were forced to be in the right-angles for those right-angle shaped edges. In fact, your method didn't tackle with it at all.
Therefore, I suggest that you should use the other word instead of 'Regularization', for example, 'Generalization', 'Modelling', or 'Refinement'
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Reviewer 3 Report
The authors responded several points. What are still insufficient include the following:
The authors have responded several points. What are still insufficient include the following:
1.GSDs represent the spatial size that a pixel in the image space corresponds and do not necessarily mean the photogrammetric metric quality, which may be influenced by the quality of orientation parameters and image point measurements, and intersection geometry. Namely, the accuracy in X,Y,Z components should be specified to justify the reference data.
2.The locations of manually measured points of building roof have not been addressed, making the justification of evaluation difficult.
3.I considered that “Last but not least, although the thresholds set showed no significant difference in terms of regularization results both quantitatively and qualitatively, the methodology in choosing optimal parameters taking different data acquisition and types of building roof into account leaves no clear guidance and more effective and sufficient corner points selection awaits actual implementation to realize the unique contribution of this study.” has not been well tackled in this version.
One suggestion for the sentences like this:
“In Figure 4 is shown an example of the results derived from critical point determination step. “ may be written as:
Figure 4 shows an example of the results derived from critical point determination step.
Author Response
Please see the attachment
Author Response File: Author Response.pdf
