Automatic Identification and Geometrical Modeling of Steel Rivets of Historical Structures from Lidar Data

Round 1

Reviewer 1 Report

This paper proposes an automatic method for rivet point extraction from TLS data, with the purpose of structure conservation. The application of this method is meaningful. The structure of this paper is reasonable and easy to read. The test design and analysis are both reasonable. My main concern is about the method design of center point estimation. Detailed comments are as follows:

 

My main focus about the paper is about Section 2.1.3: Why not firstly fit a model to rivet points and then calculate the center based on the model. Many centers or central lines of points are actually generated from a points-fitted model instead of directly from points. The center can be continuously refined in a fitting model such as Least square fitting. A model-based center estimation should be more stable, considering the uncertainty of point density. This problem has to be explained clearly, or the theoretical design of this paper seems not convincing.

 

Line82-84: However, for scan angles far from 90º (especially for angles smaller than 45º), the shape of the point cloud of the rivet base is not circular and therefore searching using a circle does not work correctly.

The description is a little confusing. Could you explain the meaning of scan angles? Does it mean incident angle? It is mentioned the shape of point cloud will change if the scan angle is too small. But laser scanning is not like photography, the shape should not change with scanning angle. The scanning angle only affects point density. So it makes me confusing and could you make a further explanation.

 

Line105-107: Departing from the obtained point cloud, a region of interest (ROI) that contains the surface of the joint where the position of the rivets is to be detected (Figure 1) is selected.

The rivets are automatically extracted from a manually selected area. It means the method is actually affected by other objects and cannot directly extract rivets from the whole scanning scene. The extraction result partially relies on the selection of ROI. So I suggest the authors to the rules of selecting ROI and make it clear under what case the proposed method may be affected by other objects/noises. For example, can this method extract rivets on a complete bridge point cloud directly?

 

Line222-225: this purpose, the average density ?? (points per rivet) of each rivet is taken as a reference. Next, a density point cloud threshold value ?? is established, where ?? = ?1 × ??, where ?1 is a constant value obtained experimentally (?1 = 1/50 from laboratory testing and field

experiments).

How to calculate ?_??

 

The colors in Figure13.d and Figure14 should be explained.

 

Is Table.1 actually a figure? I suggest it should be converted into a standard table.

Author Response

The responses to the reviewer are in the attached file.

Author Response File: Author Response.docx

Reviewer 2 Report

The authors describe a workflow to automatically detect rivets in steel structures. The main contribution is a correction of center points according to point cloud density. The approach might be relevant for practical use. But from an algorithmic point of view, it is quite simple.

Line 73: There should not be a new paragraphs because the sentence relates to the previously mentioned work of Xie et. al. 

Line 111: What can be seen at the bottom of the image?

Line 126: What is the influence of the rivets on the position of the plain. I would assume that they cause a small displacement? Why is the plane called F-F?

Indices should be indicated by a subscript (e.g, i in xi).

Line 165: How is D_k computed? Is there an influence by the shape of the rivets border?

Line 246: Why does L_j describe the shift to the real centre point? Is this a heuristics? I think the correction by  L_j is the main idea of the paper. Whereas the direction of the correction shift is obvious, its actual size should be explained. This is the main issue that I have with this paper.
 
Line 331: Only absolute values are given. It is not clear if there is a systematical error. It could be that L_j is always too large or too small?

Author Response

The responses to the reviewer are in the attached file

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

Maybe you can add your explanation of hyperplanar fitting to the paper?

line 172: indices should be indicated with a subscript, please check the paper again

line 339: It would be better if the sign of errors would be visible.

Author Response

The authors sincerely thank the great effort made by the reviewer and their constructive comments which, definitively, have allowed us to detect the weak parts of the work and improved them. The manuscript has been updated according to the reviewer's comments.

 

-Maybe you can add your explanation of hyperplanar fitting to the paper?

The explanation of hyperplanar fitting has been added to the paper.

“Initially, the plane is calculated using all points, leaving the plane slightly deviated in the direction of the rivets. Next, the points that are located at a distance greater than 0.5 times the distance from the farthest point are identified (which will coincide with a rivet head). Once these points that will belong to the rivets have been identified, the position of the plane is re-calculated without including them in the regression procedure, thus calculating a second plane that is better adjusted than the first one. This process is carried out two more times, reducing in each step the threshold distance from which points are discarded for the subsequent adjustment of the plane. The threshold distance considered in the last step results from applying a distance equal to two times the laser scanner error (r)”

 

-line 172: indices should be indicated with a subscript, please check the paper again.

The indices were indicated with a subscript.

 

- line 339: It would be better if the sign of errors would be visible.

We apologize for the poor description of Section 3.2 in the previous version of the manuscript. We have rewritten this part to clearly reflect the obtained results. It has been indicated that the error is defined as the Euclidean distance between the real position of the centre of the rivets and the one derived from the application of the proposed algorithm. As we indicated in the previous round of reviews, this error is not systematic and although it is usually in the scanning direction, it may also take place in other directions with respect to the rivet centre. The Cartesian coordinates of both points (real centre and the one derived from the application of the proposed algorithm) are referred to one corner of the corresponding model (plate of 215 × 50 mm with four 25-mm diameter rivets, plate of 215 × 70 mm with three 32-mm diameter rivets or plate of 215 × 70 mm with three 40-mm diameter rivets). This had already pointed out in the manuscript (see now lines 336 and 349), but we recognise that definitely it was not clear enough. We hope that now this issue has been successfully amended.

Author Response File: Author Response.docx