4D: A Real-Time Driver Drowsiness Detector Using Deep Learning

Round 1

Reviewer 1 Report

This paper aims to design a model that detects eye opening and closing and uses it to detect a period of eye closing as a fatigue event. The proposed model was not trained or tested in an actual or simulated driving scenario.

[line197] What is the reason for designing the 4d model this way? Besides VGG, Is there any comparison with other neural network classification models like ResNets?

[Figure 13] What do false predictions look like? In which cases does the 4d model perform better than VGGs?

[line296] How is the training done? Are the parameters trained above applied? What accuracy does the model achieve?

[line300] How is the blink rate calculated? How does the accuracy of the model affect the calculation of blink frequency? Or is it just detecting 2s of eye closure as a fatigue event, as stated in line 316?

Many details are missing. The proposed method is far-fetched to actual drowsiness-driving scenarios. The experimental design must be improved. For instance, the drowsiness detection approaches mentioned in [1] should be compared with your method. 

[1]  Yang C, Yang Z, Li W, See J. FatigueView: A Multi-Camera Video Dataset for Vision-Based Drowsiness Detection. IEEE Transactions on Intelligent Transportation Systems. 2022 Oct 27.

Author Response

Thanks for your suggestion. Please find the attached file.

Author Response File: Author Response.pdf

Reviewer 2 Report

In this paper, the authors proposed a deep neural network approach, called the deep driver drowsiness detector (4D). This approach aims to detect drowsiness of drivers. The approach proposed in this work is based on the use of learned features from convolutional neural networks in order to collect numerous facial features and other non-linear features. The MRL Eye dataset is used to train the model.

The topic of this paper is fine. However, The contribution of the paper is not clear; the idea does not seem to be novel (In the current version of the paper). It lacks clarification and analysis. The authors need to show significant improvement over other approaches to claim the consistency and the novelty of their proposal. There are some points that must be taken to improve the quality of the article:

1- The authors listed some common problems in this area of research. Since no literature review and analysis is conducted yet, it is not clear how these problems are proposed, and it is not clear if these are the problems the authors try to solve 2- The motivation of this paper should be introduced in more details. In the current, it is difficult for the readers to understand the necessity and novelty of the proposed architecture. 3- The authors should clearly point out the major contributions of this paper by using 2 to 3 brief bullet points, at the end of Section 1 (Introduction)..

4- Extensive simulations have been carried out to validate the effectiveness of the proposed method, and many experimental results are shown in the paper. However, few result analyses are given in the paper to explain the reason why the proposed approach can achieve better performance.

5- The results parts are poorly explained and discussed. it is recommended to rewrite this part.

6- I suggest that the authors provide more effort when analysing, describing and discussing the results of Figures 9, 10, 11 and 12.

7- In section "4.3. Real time implementation of drowsiness detection" I can't find any results concerning the temporal aspect of the proposed solution??? What is new for the readers in this section? what do you mean by Real time? what is the temporal constraint to consider your solution as real time?

8- The comparison results presented in table 4 are not very clear. Indeed, the author compares his results with other studies 28, 29, 3 to 31 which are not detailed and analyzed in introduction. Do they use the same MRL database?

9- . The abstract and conclusion should be rewritten to concentrate on the proposed problems and its solutions.

 Overall, the current version of this paper needs further improvements before published as a journal paper

Author Response

Thanks for your suggestion. Please find the attached file.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Most of my concerns are properly covered, and the paper can be accepted in its current form.