Transport Infrastructure Management Based on LiDAR Synthetic Data: A Deep Learning Approach with a ROADSENSE Simulator
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThe paper is interesting and suits the aim of the journal. Some comments for the authors:
Please elaborate further on the objective statement. What are the original aspects and highlight the contribution to the existing state-of-the art.
What are the current limitations and future prospects of LiDAR in transport infrastructure management?
Please highlight the way on how LiDAR can be integrated with other existing and more traditional monitoring systems in the framework of infrastructure management. Please provide (if able) an implementation framework.
Comments on the Quality of English LanguageModerate checks are needed.
Author Response
We would like to thank the reviewer for the kindness to perform this review and the insightful comments raised.
The paper is interesting and suits the aim of the journal. Some comments for the authors:
Please elaborate further on the objective statement. What are the original aspects and highlight the contribution to the existing state-of-the art.
The final paragraph of the introduction was rewritten to clearly state the objectives of the work and highlighting the contributions of the paper. The final text is as follows:
The main objective of this research work consists of designing and developing a synthetic 3D scenario and Point Cloud data generator that provides cost-effective la-belled datasets of roadside environments to be used as an input for DL semantic classifiers. The assessment of the simulator is achieved by using the obtained datasets to train the well-known CNN Pointnet++ and comparing the classification results with those accomplished with the established simulator HELIOS++. The main contributions to the stat-of-the-art consist of helping road managers to reduce the time-consuming and demanding work to achieve labelled 3D Point Cloud datasets to train DL classifiers to support in elaborating an accurate road inventory for improved safety and asset management.
What are the current limitations and future prospects of LiDAR in transport infrastructure management?
Thank you for this comment, we think is valuable and contributes to improve the vision of the paper. The following paragraph was included to the introduction focusing on the topic of LiDAR for TI management:
It is important to highlight the applicability of LiDAR, especially for infrastructure inventory and management based on semantic segmentation [4–6], where it is crucial to overcome the current limitations of the technology related to the demanding costs and workforce linked to the huge datasets obtained.
Please highlight the way on how LiDAR can be integrated with other existing and more traditional monitoring systems in the framework of infrastructure management. Please provide (if able) an implementation framework.
Thank you again for this comment, we agree that contrasting the contribution of LiDAR with the traditional workflow would improve the manuscript. Following reviewers’ recommendation and taking into consideration the previous comment also, we added a paragraph to the methodology section that address this issue as follows:
Including 3D point clouds of roads to traditional surveying methods help road managers to directly characterize infrastructure from their geometric features, permitting to perform safety assessments, planning road interventions and simulating the effects of retrofitting activities.
Comments on the Quality of English Language:
Moderate checks are needed.
We would like to thank again the reviewer to help to identify areas of improvement for the manuscript.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsThe manuscript submitted presents research classification of LiDAR point clouds trained on synthetic data. Although from the title, the manuscript seems to focus on the classification of lidar points for road inventory or similar purposes, there are pieces of text that deal with other issues related to forest and wilderness conservation. The design of the study seems adequate.
English language is, in general, correct.
Literature review contains a few references that are not necessary because they are not closely related to the topic.
Lines 31-55. I do not see such a clear connection between “climate emergency”, infrastructure vulnerability and the research presented. The introduction should be more relevant to the topic and this content could be therefore simply deleted including the counterpart references.
Line 124-140. The explanation about what roadsense is and how it works could be moved to the next section, since it details materials and methods.
Line 150 and 153: “geometric” design. In any case, section 2.1 title
Line 154 to 157: Fluff content not related to the research topic. Please, keep it relevant. It should be deleted.
Figure 1. The image is skewed.
Line 198-201. Figure caption is too large. This content should be moved to the main text. Can the simulated point cloud can incorporate these features?
Line 270. “Case study” might be a more adequate section title.
Lines 290-295. Text not relevant. Illegitimate self-citations must not be included in the text.
Line 321-346. As I mentioned above, the manuscript is scattered with some parts related to forest conservation that apparently are not closely related to the topic. Trees by the roadside might be relevant when they interfere with traffic operation. Check out the following article.
Iglesias, L.; De Santos-Berbel, C.; Pascual, V.; Castro, M. Using Small Unmanned Aerial Vehicle in 3D Modeling of Highways with Tree-Covered Roadsides to Estimate Sight Distance. Remote Sens. 2019, 11, 2625. https://doi.org/10.3390/rs11222625
Line 353: insert the link as a formatted reference.
Line 401: “creates”
Line 414-415. Sentence not clear.
Line 424-426. Table caption is too large. Move to main text. What does point features column mean?
Line 429. Please explain in the main text more about Pointnet++ and the structure of layers used, particularly in relation to figures 8, 9 and 10.
Lines 457-460. Figure caption too large. Move final clarifications to main text.
Figures 14, 15 and 16. Figure caption too large. Move final clarifications to main text.
Concerning the conclusions, beyond the performance of the algorithms utilized, I believe that the study is interesting from a road engineering point of view as the classification of lidar points opens the doors to the elaboration of high-performance road inventories and the evaluation of road safety features. This can be highlighted in the text.
I insist that the fit of the content on forests is ancillary as it is set out in the text. A more practical approach should be given to this from the road point of view as the simulated point cloud scenarios occur from the road into the environment. For example, study the impact on traffic operation or road safety.
Comments on the Quality of English LanguageSee comments above
Author Response
Comments and Suggestions for Authors:
The manuscript submitted presents research classification of LiDAR point clouds trained on synthetic data. Although from the title, the manuscript seems to focus on the classification of lidar points for road inventory or similar purposes, there are pieces of text that deal with other issues related to forest and wilderness conservation. The design of the study seems adequate.
We would like to thank the reviewer for the kindness and assistance to improve the manuscript and the helpful comments.
English language is, in general, correct.
Literature review contains a few references that are not necessary because they are not closely related to the topic.
Lines 31-55. I do not see such a clear connection between “climate emergency”, infrastructure vulnerability and the research presented. The introduction should be more relevant to the topic and this content could be therefore simply deleted including the counterpart references.
We wanted to provide a context for the manuscript in accordance with the activities of the project where the framework research project, but we agree that perhaps is a too general description and consequently this paragraph was deleted from the final text.
Line 124-140. The explanation about what roadsense is and how it works could be moved to the next section, since it details materials and methods.
Thank you for this comment, we agree that the flow of the manuscript will improve with the change. Accordingly we removed the paragraph from the introduction and moved it to methodology, reducing the redundant elements and giving an answer to the comments of the reviewers. The final text states as follows:
ROADSENSE (Road and Scenic Environment Simulation) is a novel 3D scene simulator that generates synthetic scenarios and data entirely from scratch to serve as input for DL-based semantic classifiers in road environments, including roadside forest areas. Including 3D point clouds of roads to traditional surveying methods help road managers to directly characterize infrastructure from the geometric features, permitting to perform safety assessments, planning road interventions and simulating the effects of retrofitting activities.
Line 150 and 153: “geometric” design. In any case, section 2.1 title
We would like to thank the reviewer to notice this, we changed all the references to geometric features and maintain “geometrical” only for procedures and calculations related to geometry.
Line 154 to 157: Fluff content not related to the research topic. Please, keep it relevant. It should be deleted.
We agree with the reviewer that conciseness is crucial so the paragraph was accordingly removed.
Figure 1. The image is skewed.
The image was changed to improve the significance and make it clear that we included banking cross-slope in the design of the synthetic road model.
Line 198-201. Figure caption is too large. This content should be moved to the main text. Can the simulated point cloud can incorporate these features?
We agree that the caption text is significantly large and moved it to the main text to improve readability. Also, we changed the expression of the text to show how simulated point cloud includes these characteristics. The paragraph is now written as follows:
Synthetic point clouds may differ from real ones in certain shapes like scanner occlusions. To describe this issue, Figure 2 shows a sample of a real MLS point cloud, where the green line represents the scanner trajectory and red rectangles with same sizes are shown to provide information about the point density in different parts of the point cloud referred to the MLS trajectory. For instance, rectangles 1, 2 and 3 contain respectively 25000, a few tens and approximately 500 points. Neural networks are expected to generalize and overcome these errors unless overtrained on specific features and, thus, we try to include these features by introducing selective downsampling, a random approach simulating occlusions to enhance analytical results.
Line 270. “Case study” might be a more adequate section title.
We agree with the reviewer and consequently changed the titles of section 3 and, accordingly, subsection 3.1.
Lines 290-295. Text not relevant. Illegitimate self-citations must not be included in the text.
As we mentioned in previous comments about the introduction section, we present this manuscript as a result in the context of a research project related to forest wildfires in roadsides, but we agree that the citation could be considered as out-of-scope and was removed to avoid misunderstandings.
Line 321-346. As I mentioned above, the manuscript is scattered with some parts related to forest conservation that apparently are not closely related to the topic. Trees by the roadside might be relevant when they interfere with traffic operation. Check out the following article.
Iglesias, L.; De Santos-Berbel, C.; Pascual, V.; Castro, M. Using Small Unmanned Aerial Vehicle in 3D Modeling of Highways with Tree-Covered Roadsides to Estimate Sight Distance. Remote Sens. 2019, 11, 2625. https://doi.org/10.3390/rs11222625
As we mentioned in previous comments, the research context of the paper includes the analysis of wildfire risks nearby road infrastructure, and we find very interesting to discuss the application of the simulator to road safety analysis.
The following text was added to the introduction section:
3D model analysis can also be used to detect road safety issues related to sight distance on sharp vertical curves and horizontal curves. These visibility limitations can create accident-prone areas and may require corrective measures, such as reducing the permitted speed.
The following paragraph was added to the discussion:
These results show that ROADSENSE along with HELIOS++ can help improve road management. An application use case of the simulator is related to road safety. Previous works have shown that the use of unmanned aerial vehicles (UAVs) for data acquisition on road sections that could present road safety problems is straightforward [7]. Even though UAVs make it easier for highway managers to collect data soon after identifying safety issues, national- and European-level regulations restrict flight operations over roads. ROADSENSE can generate 3D scenarios to test analytic procedures suitable for assessing sight distance on a road section. The resulting methods could be applied to actual 3D models of the road derived from the data captured by the UAV platform.
And, finally, the following paragraph was added to the conclusions:
The paper addressed how ROADSENSE can benefit road safety assessments by generating data that is suitable for road managers to evaluate the visibility distance on a road section or the vegetation condition nearby roads. This generated data includes labelling of the assets at a negligible cost compared to other sources and, what is most important, would permit to setup safety-related methodologies to be applied to actual datasets.
Line 353: insert the link as a formatted reference.
A new reference with the link to the Dataset was added to the manuscript.
Line 401: “creates”
We would like to thank the reviewer that identified the typo, that was corrected in the manuscript.
Line 414-415. Sentence not clear.
Following reviewer’s recommendation, we re-wrote the sentence to improve readability as follows:
As the number of road assets considered in this work exceed the number of tree species, the variability of the synthetic road point clouds is higher than that of nearby forest point clouds.
Line 424-426. Table caption is too large. Move to main text. What does point features column mean?
We agree that the caption text is large and moved it to the main text to improve readability.
Line 429. Please explain in the main text more about Pointnet++ and the structure of layers used, particularly in relation to figures 8, 9 and 10.
We have detailed the description of the architecture of Pointnet++ in the text in accordance with the comment. We moved the following text:
PointNet++ selects randomly some regions of the point cloud and applies PointNet in them in order to learn global and, also, local features from the point cloud.
And now states as follows:
The architecture of PointNet++ includes several key layers related to sampling, feature extraction, grouping, and segmentation, which is based on PointNet. Sampling layer is focused on efficiency thus selecting a subset of points from the input. These small sub-sets of points are used to extract features related to the fine detail of the objects. The following layer is a grouping that constructs local region sets using the Farthest Point Sampling algorithm to obtain higher-level features. The final layer based on PointNet performs the classification based on feature aggregation.
Accordingly, Table 3 caption was changed for readability:
Key properties of the default architecture of PointNet++. Each layer receives 32 sample points and their associated tensors, which contain the features to be used for the classification of the input cloud. N_Points, Radius and N_samples columns show the different sizes of PointNet characteristics for each type (ID) of layer.
Lines 457-460. Figure caption too large. Move final clarifications to main text.
We agree that the caption text is large and moved it to the main text to improve readability. The new caption is:
Figure 8. Results for training and validation of PointNet++ in highway environments with ROAD-SENSE point clouds as input data.
Figure 9. Training and validation of PointNet++ in national road environments with ROAD-SENSE point clouds as input data.
Figure 10. Training and validation of PointNet++ in forest environments with ROAD-SENSE point clouds as input data.
And the text includes now the following paragraph
These figures present the results for training and validation of PointNet++ in different environments using synthetic point clouds generated by ROADSENSE as input data. The figure includes the results using (a) the default architecture of PointNet, (b) 4 PointNet-like modified layers, (c) 5 PointNet-like layers and (d) 6 PointNet-like layers. The Overall Accuracy (OA) is depicted in blue, whereas mean accuracy (MA) is depicted in yellow. Other metrics as mean loss (ML) and mean intersection over union (MIoU) are presented in red and green respectively. The context of the results are as follows: figure 8 is focused on highway environment, figure 9 is focused on national roads and figure 10 in forest environment.
And for figures 11-13:
Figures show the resulting metrics with the same color code as in the previous ROADSENSE case, focusing also in highway (figure 11), national road (figure 12) and forest environment (figure 13).
Figures 14, 15 and 16. Figure caption too large. Move final clarifications to main text.
We agree that the caption text is large and moved it to the main text to improve readability. The text included is the following:
Figures 14 and 15, show respectively, the segmentation results for highway and national roads, with a similar color scheme is the following: black for circulation points, light grey for barriers, blue for signals, brown for DTM, orange for berms, dark grey for refuge island and green for vegetation. For forest environments, the color scheme con-sist of green and brown colors to represent vegetation and DTM points respectively.
Concerning the conclusions, beyond the performance of the algorithms utilized, I believe that the study is interesting from a road engineering point of view as the classification of lidar points opens the doors to the elaboration of high-performance road inventories and the evaluation of road safety features. This can be highlighted in the text.
I insist that the fit of the content on forests is ancillary as it is set out in the text. A more practical approach should be given to this from the road point of view as the simulated point cloud scenarios occur from the road into the environment. For example, study the impact on traffic operation or road safety.
We thank the reviewer for this insightful comment that helps to improve the manuscript and the significance of the research. Accordingly, we added the aforementioned paragraph to the conclusions:
The paper addressed how ROADSENSE can benefit road safety assessments by generating data that is suitable for road managers to evaluate the visibility distance on a road section or the vegetation condition nearby roads. This generated data includes labelling of the assets at a negligible cost compared to other sources and, what is most important, would permit to setup safety-related methodologies to be applied to actual datasets.
Comments on the Quality of English Language:
See comments above.
We take this opportunity to thank the reviewer again for the valuable comments that we gladly addressed to improve the manuscript.
Author Response File: Author Response.pdf
