Enhancing Front-Vehicle Detection in Large Vehicle Fleet Management
Round 1
Reviewer 1 Report
The paper aims to use YOLOv4 neural network for vehicle detection. However, the paper is poorly written, with many English mistakes and senseless sentences, what make it difficult to properly judge the paper merit. In my opinion, the paper must be revised and proofread before the next round of revision, so we can focus in the content.
Author Response
Dear Reviewer:
We have revised the manuscript based on your suggestions.
Please see the attachment.
Best Regards,
Ching-Yun Mu
Author Response File: Author Response.pdf
Reviewer 2 Report
The article entitled “Enhancing Front-Vehicle Detection in Large Vehicle Fleet Management”, proposes a deep learning method, as a solution for alerting large vehicle drivers to avoid serious collisions and reduce wastes of social resources and invisible carbon emission.
In particular, this research work proposes an updated version of the You Only Look Once (YOLO) approach along with a fence method based on the concept of spatial polygon for studying and analysing the front-vehicle detection.
The methodology which is followed in the proposed approach is clearly defined and supported by adequate experiments permitting other researchers to reproduce certain aspects of their results. Additionally, the methodology analysis, as well as the results are enriched with an efficient number of properly presented tables, figures and charts.
Nevertheless, although the article is utterly interesting, the authors should pay some attention in the following issues:
- Although the Introduction Section as well as the second section “Related Works” both provide a satisfactory number of references, the novelty as well as the contribution to knowledge of the proposed solution is not clearly substantiated. It is strongly suggested that the authors should take care of this issue and put more effort in clarifying the research question as well as the motivation for conducting this research. In addition, the authors should provide, for the benefit of the readership, more details on the YOLO approach (earlier and latter versions).
- Although the paper is well-structured and in general written in appropriate English language according to the standards of the Journal, more attention should be put on the meaning of some sentences in order to be more understandable by readers.
Author Response
Dear Reviewer:
We have revised the manuscript based on your suggestions.
Please see the attachment.
Best Regards,
Ching-Yun Mu
Author Response File: Author Response.pdf
Reviewer 3 Report
Dear Authors,
Many thanks for your manuscript submission to MDPI Journal of Remote Sensing. After careful review, I justify that in general, this research article presents comprehensively good set of work on front-vehicle detection by combining approaches on image enhancement, machine learning, remote surveillance and object classification, while the technical schemes compared Harr feature selection, YOLOv4 and its variations. The use of English is just acceptable, the organization of this paper is also fairly good. There are a few aspects need improvement before coming to the recommendation of minor revision towards acceptnace, which I enumerated (may not limited to those) as follows:
Major problematic issues suggested for improvement in your revision:
a) Abstract: I think it is generally in good shape, while a better version is one condensed to within 180~200 words in total. For instance, narrations in Lines 9-14 can be condensed, and the last sentence also needs shortening. Please consider the suggested edits for reference. Some keynote quantitative results, are supposed to be included. Thanks a lot!
b) The Introduction Section: Some paragraphs may require a major rewrite on this section; the review of state-of-the-art models missed any prior work in Years 2010-2015; also, if you citing more latest publications, they should be included in this section; meanwhile, I think the authors missed a main summary of major contributions (3-4 manifolds), hence, updating this part with more specific details, would possibly be a better option.
c) Section 2: Good observation on related work. I suggest the authors present the review of single-stage and two-stage vehicle detection schemes in a separate subsection. The CNN-based approaches may already have a lot of variations due to the fast progress of deep learning, which could be taken into some account. YOLO v5 has also presented its model and variations, which should be included in this section. The authors need to compare and state the advantages and shortcomings between two-stage and one-stage approaches on vehicle detection, then address the importance and reason of why selecting YOLOv4 as their major approach. Thanks very much!
d) The Section 3 looks like Related work on summarizing the general flow of YOLOv4 and its related technical schemes. What is the difference of your scheme in contrast to pure Haar and original YOLOv4? What is the core metric on machine learning related performance evaluation (i.e., loss function)? Can you update this part with more specific descriptions? PS: Another possible defect (or potential question mark) is that, the current version lacks professional formulas for Evaluation (just basic information retrieval metrics, while the size of each equation should be zoomed in), and the section seems missing mathematical derivations; if any, please apply edits in your updated version. Thanks a lot!
e) Some issues for improvements on Figures 6 and 8: the authors had better use larger size of images, larger (uniform) style of characters (i.e., Times New Roman, size = 12) for all the legends; besides, the size, resolution of each figure should be calibrated to standard size similar to MDPI template.
f) Section 4 (Results): First, supplement the formulas (if any) and the enhanced evaluations metrics (if any), which may include the classifed number of counts, loss function, and enhanced measures established in Years 2014-2018. Explain the differences among the three performance matrics (elapsed time, FPS and variances in Tables 7-9, respectively) between Haar and YOLOv4(III).
g) Section 5 (Discussion): a bit too generic. In addtion to one single limitation of your whole paper, discussions should include your justification on the limitations of study (with specific statements) along with some of the quantitative results. Also, if any ablation study or sensitivity analysis, those can be included in the expanded section.
h) Conclusion Section: I suggest that a second paragraph can be inserted (representing discussions on opening problems and your limitations on study), besides, the concluding remarks should be supplemented with some statements on summary of research challenges and a few specific future orientations on study.
i) References: 1) I think the authors may had neglected on citing highly visible ACM/IEEE Transactions for machine learning based front vehicle classification and fall detection (typically in Years 2016-2021); 2) while the current shape take good care of matching the template on MDPI Journal, citations on some conference proceedings missed time, locations and page range, please just apply the updates; 3) Check and add the related citations in close research topics on parallel comparison (Years 2010-2015,2019-2022), i.e., MDPI Journal of Remote Sensing / Sensors / Applied Science / Entropy / Electronics, then justify whether any matched work are good matches on citing; 4) keep the citation style uniformly consistent (abbreviation, italic, bold, full-stop and the end) of each reference. Thanks very much!
Minor issues suggested to be calibrated in the revised version:
a) If you use MS word or Latex, please avoid hyphenating a word (which currently appears many times at the end of some lines to cross-over two adjacent lines). Refer to your MDPI online template, which may have the options to adjust that issue. Thanks a lot!
b) Literal quality of English writing is acceptable, while there is still room for improvement in the updated version. A few minor typing and grammatical issue need to be fixed, for instance, I think the "t-value" and "p-value" should be use italic while lower-case letter; be sure the half-space interval are fine among words between sentences. Meanwhile, please proof reading the related context carefully.
c) Align the size of figures, tables, and the interval before and after therein.
Once again, thank you for the manuscript submission to MDPI Journal. We appreciate your future efforts and wish you good luck to improve the overall quality of your research work. All the best for your paper acceptance!
Stay safe,
With warm regards,
Author Response
Dear Reviewer:
We have revised the manuscript based on your suggestions.
Please see the attachment.
Best Regards,
Ching-Yun Mu
Author Response File: Author Response.pdf
Reviewer 4 Report
- Image processing methods in recent years mainly use artificial intelligence, including artificial neural networks. One of the most common applications of object detection methods in images are object detection based on video data. This article presents one of the problems related to the detection based on a video data for front vehicle detection.
- General remarks
- Too many abbreviations make it difficult to follow the content of the article. Each abbreviation should be expanded the first time it appears. Not all readers need to know all abbreviations. Especially in the abstract of the article and conclusions. Also “Haar” should be descripted and explained.
- Please use the language of a scientific research report without personal references like “we” and “our” included “our own” which are used many times in whole article.
- The paper lacks more precision information about the camera used during experiment.
- However the article is well written should be carefully edited. Some remarks included below.
- Specific remarks
- Line 234 and 285 “you only look once” was explained in line 221. Please erase it.
- Line 446 - what means “rpart and partykit“?
- The final conclusions are too general and only generally summarize the research presented in the article. I suggest expanding the conclusions with more detailed findings.
Author Response
Dear Reviewer:
We have revised the manuscript based on your suggestions.
Please see the attachment.
Best Regards,
Ching-Yun Mu
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
The paper proposes the use of YoloV4 for front rear detection from videos. The paper has considerable improved from the previous version. The approach is interesting, but I have some remarks for improving the paper:
- It would be interesting to better describe the content of the "seven documents" mentioned in Figure1. They may understood by someone who understand the tools you have used, but for the sake of readability, this should be better described.
- The same applied to Figure 2, and a better description should be provided.
- The formulas and contingency matrix are well known in the literature, and in my opinion could be removed from the paper and referenced properly.
- The baseline method used (decision tree) is quite simple. In my opinion, it would be interesting to add an extra based (e.g., Random Forest) over the Haar features.
- Tables 3 and 5 are connected? The numbers seems not to match.
- Finally, the section 5 (Discussion) could be improved.
Author Response
Dear Reviewer:
We have revised the manuscript based on your suggestions.
Please see the attachment.
Best Regards,
Ching-Yun Mu
Author Response File: Author Response.pdf