Real-Time Personal Protective Equipment Compliance Detection Based on Deep Learning Algorithm

Round 1

Reviewer 1 Report

 

Deep learning algorithms in construction industry for detecting workers without PPE can help reduce many dangerous accidents in injuries. The paper is a well-structured and summarized. The experimental validations are convincing. The paper can be accepted, subject to doing the following major corrections.

1.The novelty of the proposed approach remains rather unclear. It is stated that two algorithms, YOLOv3 and YOLOv4, are applied. The contributions regarding these should be elaborated.

2.     The previous approaches and their relation to the presented method should be elaborated. The contributions of the manuscript should be stated in much more detail. In Introduction, the problem addressed in the manuscript should be described more clearly and in more detail. The actual introduction to the topic of the manuscript is short, as most of Introduction is actually discussing related works. These papers may also give the authors more idea on various object detection techniques that can be explored and included in the introduction part.

https://doi.org/10.1007/s00542-019-04694-8

https://doi.org/10.3390/math6100213

https://doi.org/10.1016/j.ecoinf.2021.101469

https://doi.org/10.1016/j.procs.2022.01.135

3.     The authors can include more details about the preprocessing techniques used. The selection of the components of the proposed method are not justified adequately.

4.     The proposed block diagrams can be drawn with more clarity and legibility of labels. The bounding boxes where ever drawn (say in Figures 1, 4 and 9) can be drawn with contrasting colours for better visibility. It would be better to use vector format for the Figures as far as possible.

5.     In YOLOv3 the usage of down sampling strides is not clear. Explain it for readers understanding.

6.     Authors can include more graphs depicting the performance of their proposed methods. Inferences can be included for all the graphs in terms of performance measures.

7.     The authors are recommended to include complexity comparison between the techniques used with sufficient discussions.

8.     Figure 4, 8 and few others contain the picture of a human. Is any ethical clearance obtained for it? Or any other relevant document that justify your statement.

9.     The experiments in the manuscript are not fully convincing and should be extended in a major way. The actual task in the experiments is not clear as sometimes it is regarded as object tracking and sometimes as object recognition. The reported results should be positioned to the state-of-the-art. Preferably, the experiments should include some standard benchmarks as well. Overall with the current experiments, it is difficult to assess the performance of the proposed method in relation to the state-of-the-art. The authors should prove that their obtained results are good compared to state of art techniques and this paper  https://doi.org/10.1007/s11042-022-13281-5 for modified detection algorithms using YOLO can be used and compare their works with many other such relatable works.

10.  What are the possible limitations of this research paper if any? The authors are suggested to add a single subsection after experiments to further discuss the limitations.

 

 

 

 

 

Author Response

Thank you for inviting us to submit a revised draft of our manuscript entitled, “Real-time personal protective equipment compliance detection based on deep learning algorithm” to Sustainability. We also appreciate the time and effort you have dedicated to providing insightful feedback on ways to strengthen our paper. Thus, it is with great pleasure that we resubmit our article for further consideration. 

We have incorporated changes that reflect the detailed suggestions you have graciously provided. We also hope that our edits and the responses we provide below satisfactorily address all the issues and concerns you have noted.

To facilitate your review of our revisions, the following is a point-by-point response to the questions and comments delivered in your letter dated 11/22.

Thank you for providing these insights. All the additions and comments give us many areas where we can improve.

1. We agree with you. In order to ensure the novelty of this study, the relevant content and data analysis of the YOLO7 model has been added, please refer to line 210, Table 2, Figure 9, and Figure 12.
2. Thank you for your suggestion. The contributions of this study have been supplemented in the Introduction and added references 25, 26, and 28.
3. We agree with your assessment. The introduction of Mosaic data augmentation and mixup has been expanded, please refer to line 256 and Figure 7.

4. We agree with you. Except for Figure 5, which is the UI, the rest of the images have been improved in quality.

5. We agree with your assessment. The content has been added for the downsampling of YOLOV3, please refer to line 156.
6. Thank you for your suggestion. Some images about model performance have been added, please refer to Figure 9 and Figure 12.
7. Thank you for your suggestion. We added some comparisons between YOLOv7 and YOLOv4 please refer to line 213.
8. We did not apply for ethics clearance, all images in this study were taken by us, and Mosaic processing has been performed on some portraits.
9. We agree with your assessment. The latest work in the YOLO series has been added to this study.
10. Thank you for your suggestion. Limitations and Future Work have been added to this study, please refer to line 337.

 

Again, thank you for giving us the opportunity to strengthen our manuscript with your valuable comments and queries. We have worked hard to incorporate your feedback and hope that these revisions persuade you to accept our submission.

Author Response File: Author Response.docx

Reviewer 2 Report

The paper is devoted to various aspects that are associated with determining the availability of various personal protective equipment and safety for construction workers at the time they are at work. The authors of the paper describe in some detail the main problems of the construction industry, including the presence of a large percentage of various injuries to people at the time of performing work duties. Workers usually get such injuries due to negligent observance of all safety rules in the workplace. The authors of the paper make the main emphasis on the statement that at present there are several options for monitoring workers during construction work. The first is the visual control of a specially trained person. The second is the wearing of special sensors by workers. The third is automatic visual surveillance through video cameras. One can fully agree with the authors of the paper, who argue that the third method is the most effective compared to the first two, but only in the case of a high-quality pre-trained neural network model for recognizing and localizing personal protective equipment on construction workers. It is on the training of such neural network models that the study of the authors of the paper is directed. Testing and comparative analysis of the results of the study were carried out on the Personal Protective Equipment (PPE) dataset. On the basis of comparative analysis, it can be argued that qualitative indicators still depend on the input data. However, the paper is not free from a number of shortcomings.

 

1) First of all, the absence of a description of the selected data set and the absence of a link to the source are striking. Is the original source https://universe.roboflow.com/personal-protective-equipment/ppes-kaxsi or something else? If the original source is roboflow, then how does the markup of the authors differ from the original roboflow?

2) The authors used Yolo neural network architectures for training. The question arises why is there no description of Yolo version higher than 4? Why weren't they used for teaching?

3) There is no comparison of the obtained results with other new papers. If these results by the authors should be considered the World Baseline, then this should be mentioned in the paper and said that there are no other results for this data set.

4) Some drawings are of poor quality; they should be improved in quality.

5) There is no description of modern hybrid methods for detecting objects, which are constantly presented at conferences focused on visual modality (CVPR, ICCV, ECCV, and others).

6) Finally, the style of the paper requires minor revision due to the presence of spelling and punctuation errors.

 

It seems to me that all the proposed additions and comments will improve this paper. In general, the topic is relevant and the study has a place to be, but there are points that need to be finalized and clarified for the scientific community. Needs to be improved.

Author Response

Thank you for inviting us to submit a revised draft of our manuscript entitled, “Real-time personal protective equipment compliance detection based on deep learning algorithm” to Sustainability. We also appreciate the time and effort you have dedicated to providing insightful feedback on ways to strengthen our paper. Thus, it is with great pleasure that we resubmit our article for further consideration.

We have incorporated changes that reflect the detailed suggestions you have graciously provided. We also hope that our edits and the responses we provide below satisfactorily address all the issues and concerns you have noted.

To facilitate your review of our revisions, the following is a point-by-point response to the questions and comments delivered in your letter dated 11/22.

 

Thank you for providing these insights. All the additions and comments give us many areas where we can improve.

1. We agree with you. The PPE dataset was released for the first time in this study and provided the access method at the end of the article.
2. You have raised an important question. In order to ensure the novelty of this study, the relevant content and data analysis of the YOLO7 model have been added, please refer to line 210, Table 2, Figure 9, and Figure 12.
3. We agree with you. The statement that there are no other results for this data set has been added, please refer to line 229.
4. We agree with you. The image quality in this study has been further improved
5. Thank you for your suggestion. The latest work in the YOLO series has been added to this study.
6. We agree with you. We've made some spelling and punctuation corrections.

 

Again, thank you for giving us the opportunity to strengthen our manuscript with your valuable comments and queries. We have worked hard to incorporate your feedback and hope that these revisions persuade you to accept our submission.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

No further comments.

Reviewer 2 Report

The authors of the paper finalized the paper and made all the necessary changes. In this form, the paper can be useful and interesting for many specialists who connect their research with computer vision and machine learning.