A Multi-Feature Fusion and Situation Awareness-Based Method for Fatigue Driving Level Determination

Round 1

Reviewer 1 Report

The paper is interesting and the research topic is important. The research methodology is clear. The English language is OK, however, proofreading is advised to correct mistypes. 

The overall impression is good. Together with that, authors are advised to consider the following comments:

1) As far as the fatigue detection will lead to decision-making and possible control of the vehicle (or signalling about the fatigue state at least), the Dlib library's trustworthiness is highly questionable. Whether the open-source lib can be used for safety-related purposes? This is an open question for the automotive domain (e.g. like OSS and ASPICE) - Discussion, Conclusion sections? Authors are advised to put their considerations;

2) Considering the interconnectivity of the vehicle, where the proposed solution is foreseen to be implemented using special software and hardware platform, the cybersecurity aspect of the data transmission channel (as well as trust in the datasets, which obviously should be certified and approved due to the safety relation) and the cybersafety aspect is foreseen. Indeed, cybersecurity and safety interrelation is still challenging e.g. in the automotive domain authors are advised to think about it and put their considerations (Discussion, Conclusion sections?)

3) Provide the grounding of numbers from Table 2, levelling in Table 3;

4) Formatting of formulas 3-6 is foreseen;

5) Provide more information and grounding on the judgement matrix (line 317). A typical problem in expert-based assessment is entropy and subjectivity. How trustable the experts were? Any choice procedure? How the obtained numbers have been validated?;

5) The choice of LSTM needs to be explained better (line 371);

6) Provide the analysis of the results given in Tables 9-11;

7) Authors are advised to show the validation and verification of the results obtained. It can be challenging, but the results need to be verifiable, justifiable and trustable;

 

8) Authors are advised to change the status of the manuscript from "Technical note" to "Article" according to the available types of the publications https://www.mdpi.com/journal/electronics/instructions

The English language is OK, however, proofreading is advised to correct mistypes. 

Author Response

Thank you very much for your interest in my thesis. Your advice is very valuable to me. Based on your suggestions, I have made revisions to the original manuscript and submitted a revised version. I will respond to your suggestion with the file I sent you.

Author Response File: Author Response.pdf

Reviewer 2 Report

Dear authors,

please find the comments attached. Special attention must be paid to the formatting of the paper, a large number of technical errors and the description of the methodology used. It is suggested to include a flowchart in the introduction so that it will be much clearer for the readers of your paper later.
Yours sincerely

Comments for author File: Comments.pdf

Author Response

Thank you very much for your attention to my paper. Your suggestions are invaluable to me. Based on your recommendations, I have made revisions to the original manuscript and submitted the modified version (both with tracked changes and without tracked changes). I will provide responses to your suggestions through the document I am sending to you.

Author Response File: Author Response.pdf

Reviewer 3 Report

The manuscript presents an actual and possibly useful study on the analysis of driver fatigue detection using facial feature extraction and AHP-fuzzy evaluation model. Overall, the manuscript provides interesting insights into the topic. Please allow me to make a few suggestions for the authors to improve the manuscript.

The number of bibliographic references seems a bit small. I suggest the authors to improve it with other studies and publications in the studied field.

Please improve the resolution of some of the figures (for example Figure 1 etc.), to be better viewed by the readers of this journal, because in the pdf file received for review, some seem pixelated, stretched etc.

I suggest that the analysis of the results should be extended in its scope. The manuscript primarily presents tables and matrices, but I suggest the authors to provide more in-depth interpretations, to expand on the significance of the results and discuss their implications for driver fatigue detection. It can involve discussing some more on the statistical significance of the findings, exploring correlations between different features, and highlighting any unexpected or interesting patterns observed in the data.
I suggest the authors to clarify more the affiliation tables (Table 10 and Table 11) by providing further explanations of the affiliation standards, how they are applied to the dataset. I could make it easier for the readers of this journal to understand the analysis and results.

In order to establish the effectiveness of the proposed approach, I suggest that the authors could further compare some more their results with those obtained using other established methods for driver fatigue detection. It can include some benchmarking against some other popular methods, like perhaps machine learning algorithms, Convolutional Neural Networks (CNN), transfer learning techniques etc.

I suggest that the authors could elaborate some more on the Conclusion by summarizing the implications of their key findings, by discussing the significance of the study in the broader context of fatigue detection in drivers and also propose avenues for further investigation. I believe it is important for the authors to acknowledge the limitations of their study and propose some future directions for improvement (more than one sentence in the Conclusions). This could involve discussing potential sources of bias or error in the experimental setup and suggesting areas for further research and development.
It would be interesting to see (perhaps in the future) a continuation of this research, possibly by using:
-deep learning techniques (eg., convolutional neural networks, for analyzing facial features and predicting driver fatigue)
-further sensor-based data collection - by incorporating sensors, such as heart rate monitors or eye-tracking devices, electroencephalography (EEG), real-time monitoring systems (such as wearable devices or in-vehicle sensors etc.), to gather more comprehensive data on driver fatigue.
- real-world driving tests on driving scenarios to validate the findings from simulated driving.
- longitudinal study - by extending the research to collect data from the same drivers over an extended period.
- a larger and more diverse dataset - to ensure the generalizability of the findings and to account for individual differences in fatigue patterns etc.

I suggest a further re-reading perhaps by a native speaker with attention to phrases (where one ends, and the other begins). For example in Row 642 “… reaching 92%.there is still room for further improvement. “

Author Response

Thank you very much for your attention to my paper. Your suggestions are invaluable to me. Based on your recommendations, I have made revisions to the original manuscript and submitted the modified version (both with tracked changes and without tracked changes). I will provide responses to your suggestions through the document I am sending to you.

Author Response File: Author Response.pdf

Reviewer 4 Report

The presented study proposed AHP-fuzzy comprehensive evaluation algorithm to assess the driver's fatigue state with the Shape_Predictor_68_ face_landmarks.dat model library from the Dlib library to perform accurate detection of the 68 key points on the human face.

Future situation levels are predicted based on past and present states, with situation prediction method based on the improved Marine Predators Algorithm and optimized GRU (Gate Recurrent Unit). First, the situation evaluation result, which represents the "state"  is determined.

After, the identified indicators for fatigue state include mouth state, eye state, head pitch angle, head roll angle, and the presence of improper driving behavior are determined.  

It is concluded that the proposed improved GRU algorithm achieves significantly enhanced accuracy (92%) compared to the non-optimized GRU algorithm and by employing fusion analysis techniques, the impact of imbalanced data on model accuracy is reduced.

 

There are a lot of typos (format) like blank space, point, comma  etc. and because of that need minor revision.

Author Response

Thank you very much for your attention to my paper. Your suggestions are highly valuable to me. I have conducted a comprehensive format check on the entire manuscript and have made corrections to any errors that were identified. Based on your recommendations, I have made revisions to the original manuscript and have submitted the modified version (including both tracked changes and clean version).

Round 2

Reviewer 2 Report

I thank the authors for their kind comments.