Point-Cloud Segmentation for 3D Edge Detection and Vectorization

Round 1

Reviewer 1 Report

The article describes a set of tools and a methodology for the point cloud segmentation to extract 3d edges in the photogrammetric domain. It is well-written in a clear form presenting methods and data in a complete and reproducible way. The presence of a GitHub repository where you can download the python code is an important aspect for the transparency and rea-usability of the proposed approach. The article is of interest for the scientific community and represents an important proposal in order to connect arm tools with survey methodologies in CH domain.

Author Response

We would like to thank you for the time you spent to review our paper. Please find attached a details of the revision in .pdf format.

Author Response File: Author Response.pdf

Reviewer 2 Report

Authors propose a full pipeline for automatic vectorization of point clouds of architectural models.

The pipeline is obtained by integrating various well-known components in a script, and outputs a CAD model containing a semantic segmentation of 3D vectors.

The problem is interesting and very relevant, but the proposed solution appears still far from providing adequate results: the presented results appear to contain too many artifacts and the output drawings contain too many vectors:  simplification  and topology analysis steps are missing. 

Moreover, the proposed solution is based on well-known processing methods and does not contain significant contributions.

On the other side, community is addressing this problem by investigating deep learning strategies, and authors should consider this avenue, at least for what concerns discussion of methods and comparison of results.

For example, authors should consider the paper

Chuang, Tzu-Yi, and Cheng-Che Sung. "Learning-guided point cloud vectorization for building component modeling." Automation in Construction 132 (2021): 103978.

 Other important references not considered:

Bassier, Maarten, Maarten Vergauwen, and Bjorn Van Genechten. "Automated semantic labelling of 3D vector models for scan-to-BIM." Proceedings of the 4th Annual International Conference on Architecture and Civil Engineering (ACE2016). België, 2016.

Macher, Hélène, Tania Landes, and Pierre Grussenmeyer. "From point clouds to building information models: 3D semi-automatic reconstruction of indoors of existing buildings." Applied Sciences 7.10 (2017): 1030.

Ochmann, Sebastian, Richard Vock, and Reinhard Klein. "Automatic reconstruction of fully volumetric 3D building models from oriented point clouds." ISPRS journal of photogrammetry and remote sensing 151 (2019): 251-262.

Werbrouck, Jeroen, et al. "Scan-to-graph: Semantic enrichment of existing building geometry." Automation in Construction 119 (2020): 103286.

Author Response

We would like to thank you for the time you spent to review our paper. Additionally your suggestions were critical and we believe that they improved our effort. Please find attached a details of the revision.

Author Response File: Author Response.pdf

Reviewer 3 Report

The article is about an automated pipeline which addresses the problem of detecting 3D edges in point clouds by leveraging a set of RGB images and their 2D edge maps. It has been tested with images from an archaeological site of the ancient Kymissala and the old police station.

The topic of documentation of 3D reconstructions is highly relevant and – beside numerous attempts and proposed solutions - still not properly solved. Against this background, the problem statement of the article as automating the currently highly time and labour intense creation of 2D-3D architectural drawings is very interesting. As a minor suggestion: Concerning the potential audience of J. Heritage as general heritage scholars and professional some more concrete examples for usage of the 3D-2D line detector may be beneficial for their understanding.

With regards to a state-of-the-art analysis, references named are comprehensive, of relevance and up to date. From a formal point of view, an argumentation within the article is transparent and stringent.

Regarding to a scientific contribution, the article results from a diploma thesis and provides mainly a proof-of-concept for a 3D pipeline-based line detector based on two SfM frameworks and a codification into a separate layer. The article is coherent in the current form. As a suggestion, especially the evaluation of the pipeline could be extended by future research. E.g. it is currently missing inclusion of datasets providing ground-truth (e.g. 2D excavation / architectural plans). It may also be interesting to assess the pipeline systematically with further objects and also statistically assess e.g. systematic errors.

P.S.: Caption of Fig 12 „The proposed approach and the Temple of Demeter in Naxos dataset“ may need to be updated.

Author Response

Firstly we would like to thank you for the time you spent to review our paper. Additionally your suggestions were critical and we believe that they improved our effort. Please find attached a details of the revision in ".pdf" format.

Author Response File: Author Response.pdf