Workers’ Unsafe Actions When Working at Heights: Detecting from Images
Round 1
Reviewer 1 Report
1 The language is poor and there are a lot of typos；
2 There are five unsafe actions descripted in Tab.1, whether all cases are covered for your model?
3 Since the workers’ unsafe actions are extremely complicated in real work environment, how can you ensure the robust of the suggested model with the limited features of data set?
Author Response
Dear Reviewer,
We appreciate your kind work to this manuscript, which greatly helped us improve this research. We revised the manuscript carefully and outlined our responses to the specific comments. All the changes in the updated manuscript are highlighted in blue. The following is the point-to-point response to the comments.
Thanks very much for your kind work to this manuscript.
Author Response File: 
Author Response.docx
Reviewer 2 Report
The article focuses on a very important topic - safety on construction sites. Authors use well-established object detection techniques and adopt them to the specific use-case.
Overall the article is well structured and easy to read and understand. The practical contribution (setup of experiments) is not exactly novel but is of good quality. Taking into account that the research focuses on the detection of relatively rare unsafe actions, preparing comprehensive datasets is a challenging task. Perhaps semi-supervised machine learning methods should be investigated.
Reviewer would recommend to decrease the amount of academic theory in the article (for example explanation of 1, 2, 3 formulas) and focus more on practical aspects of the research.
Pictures of parkour jumping and lying on yoga mat seems to be completely out of context. Also it is not clear how selected unsafe action categories cohere with legit similarly looking construction actions (e.g. mounting overhead lamp might look very similar to throwing waste).
Article lacks the information about the size of prepared dataset, training and testing hardware and software setup. Comparing detection time between different classes looks strange, especially when fixed-pipeline object detection model is used.
Plotting test results (figure 8) as a single series is not correct - these are different indicators. Better option to use at least bar charts or ROC, Precision-Recall or similar curves.
Few notes:
- table 1 row separation could be improved
- purpose of 3.3 and 3.4 sections is not very clear
Author Response
Dear Reviewer,
We appreciate your kind work to this manuscript, which greatly helped us improve this research. We revised the manuscript carefully and outlined our responses to the specific comments. All the changes in the updated manuscript are highlighted in blue. The following is the point-to-point response to the comments.
Thanks very much for your kind work to this manuscript.
Author Response File: Author Response.docx
Reviewer 3 Report
It is an interesting and very important article from the point of view of the occupational health risk management.
I would like to express great appreciation to the authors of this research. This research was well planned, carried out and implemented. The proposed methodology is innovative and described in detail. The structure of the paper is correct. The results are clearly presented.The obtained research results are very interesting and valuable for practical use.
Overall, I believe that the article provides valuable content to the present body-of-knowledge.
There are several insufficiencies that need to be improved.
1. the title is interesting, although it may be a bit ambitious and too long.
2. in this paper I do not see very important section: Discussion. Unfortunately, this paper does not point out the shortcomings of past research to show the value of this research. Add a strengths and weaknesses section and limitations section of this research to the new section: Discussion. The discussion should refer to other studies, indicate the shortcomings of the research. The research has some limitations. The manuscript should highlight some of these limitations.
3. I believe that the following papers could add a significant contribution to the literature review (please consider adding position to the reference):
- Martínez-Rojas, M.; Gacto, M.J.; Vitiello, A.; Acampora, G.; Soto-Hidalgo, J.M. An Internet of Things and Fuzzy Markup Language Based Approach to Prevent the Risk of Falling Object Accidents in the Execution Phase of Construction Projects. Sensors 2021, 21, 6461. https://doi.org/10.3390/s21196461
- Mariusz Szóstak, Bożena Hoła, Paweł Bogusławski, Identification of accident scenarios involving scaffolding, Automation in Construction,
Volume 126, 2021, 103690, ISSN 0926-5805, https://doi.org/10.1016/j.autcon.2021.103690
- Kishor Bhagwat, Venkata Santosh Kumar Delhi, Prakash Nanthagopalan. (2021) Construction Safety Performance Measurement Using Leading Indicator-based Jobsite Safety Inspection Method - A Case Study of Building Construction Project. International Journal of Occupational Safety and Ergonomics 0:ja, pages 1-27
-C, Vigneshkumar, and Urmi Ravindra Salve. 2020. “A Scientometric Analysis and Review of Fall from Height Research in Construction ”. Construction Economics and Building 20 (1). https://doi.org/10.5130/AJCEB.v20i1.6802.
Author Response
Dear Reviewer,
We appreciate your kind work to this manuscript, which greatly helped us improve this research. We revised the manuscript carefully and outlined our responses to the specific comments. All the changes in the updated manuscript are highlighted in blue. The following is the point-to-point response to the comments.
Thanks very much for your kind work to this manuscript.
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
Authors revised the manuscript well. I recommend that manuscript to be published in Sustainability.
