An Autonomous Intelligent Liability Determination Method for Minor Accidents Based on Collision Detection and Large Language Models
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsAbstract
The topic of the paper is very current, interesting and useful for practical application.
Introduction
One of the biggest concerns is the very small number of cited literature, for such a current and extensive topic.
Please expand the introduction using more recent references and expand the picture of the analysed topic, e.g.
The occurrence of minor traffic accidents is significantly influenced by factors such as road conditions and human behaviour (Simić et al., 2023; Trifunović et al., 2024), as well as the increasingly critical aspects of traffic flow management (Kan et al., 2024) and regulation (Zlatkovic et al., 2023; Sheela et al., 2023). On the other hand, advancements in technology, engineering, and various computational tools can substantially predict and proactively prevent traffic accidents (Stević et al., 2022, Kodepogu et al., 2023).
Zlatkovic, M., Cvijovic, Z., Stevanovic, A., & Song, Y. (2023). Concepts of Signal Control Preemption for Emergency Vehicles in Connected Vehicle Environments. Journal of Road and Traffic Engineering, 69(2), 1-7. https://doi.org/10.31075/PIS.69.02.01
Sheela, S., Nataraj, K. R., & Mallikarjunaswamy, S. (2023). A Comprehensive Exploration of Resource Allocation Strategies within Vehicle Ad-Hoc Networks. Mechatron. Intell Transp. Syst., 2(3), 169-190. https://doi.org/10.56578/mits020305
Kan, H. Y., Li, C., & Wang, Z. Q. (2024). An Integrated Convolutional Neural Network-Bidirectional Long Short-Term Memory-Attention Mechanism Model for Enhanced Highway Traffic Flow Prediction. J. Urban Dev. Manag., 3(1), 18-33. https://doi.org/10.56578/judm030102
Trifunović, A., Senić, A., Čičević, S., Ivanišević, T., Vukšić, V., & Simović, S. (2024). Evaluating the Road Environment Through the Lens of Professional Drivers: A Traffic Safety Perspective. Mechatron. Intell Transp. Syst., 3(1), 31-38. https://doi.org/10.56578/mits030103
Stević, Ž., Subotić, M., Softić, E., & Božić, B. (2022). Multi-Criteria Decision-Making Model for Evaluating Safety of Road Sections. J. Intell. Manag. Decis., 1(2), 78-87. https://doi.org/10.56578/jimd010201
Kodepogu, K., Manjeti, V. B., & Siriki, A. B. (2023). Machine Learning for Road Accident Severity Prediction. Mechatron. Intell Transp. Syst., 2(4), 211-226. https://doi.org/10.56578/mits020403
Simić N, Ivanišević N, Nedeljković Đ, Senić A, Stojadinović Z, Ivanović M. Early Highway Construction Cost Estimation: Selection of Key Cost Drivers. Sustainability. 2023; 15(6):5584. https://doi.org/10.3390/su15065584
These references are my suggestions, you can use more references than this suggested to enrich the introduction with similar research in the world.
Collision Detection and Multimodal Large Language Model
What is the accuracy rate of the vehicle recognition and collision detection components in diverse environmental conditions?
What metrics were used to evaluate the accuracy and efficiency of the proposed method? How do these metrics compare to existing methods?
Are there any known limitations or potential challenges in implementing this method in real-world traffic systems?
How does the system handle cases with incomplete or ambiguous data, such as missing frames or occluded vehicles?
Conclusion and Outlook
In conclusion, you can write that this method could also have a preventive effect on reducing traffic accidents, especially the increasingly frequent "intentional" traffic accidents (for the purpose of fraud) (Arciniegas-Ayala et al., 2024; Muktar and Fono, 2024).
Muktar, B.; Fono, V. Toward Safer Roads: Predicting the Severity of Traffic Accidents in Montreal Using Machine Learning. Electronics 2024, 13, 3036. https://doi.org/10.3390/electronics13153036
Arciniegas-Ayala, C.; Marcillo, P.; Valdivieso Caraguay, Á.L.; Hernández-Álvarez, M. Prediction of Accident Risk Levels in Traffic Accidents Using Deep Learning and Radial Basis Function Neural Networks Applied to a Dataset with Information on Driving Events. Appl. Sci. 2024, 14, 6248. https://doi.org/10.3390/app14146248
The conclusion missing the future directions of research, the limitations of the study and the practical application of the obtained results.
Author Response
Dear Reviewer,
Thank you very much for your detailed review and valuable suggestions. I have carefully reviewed your comments and made the necessary revisions to my paper. Below is my response to each of your comments, including the specific locations of the changes made:
Comments 1: One of the biggest concerns is the very small number of cited literature, for such a current and extensive topic. Please expand the introduction using more recent references and expand the picture of the analyzed topic.
The occurrence of minor traffic accidents is significantly influenced by factors such as road conditions and human behaviour (Simić et al., 2023; Trifunović et al., 2024), as well as the increasingly critical aspects of traffic flow management (Kan et al., 2024) and regulation (Zlatkovic et al., 2023; Sheela et al., 2023). On the other hand, advancements in technology, engineering, and various computational tools can substantially predict and proactively prevent traffic accidents (Stević et al., 2022, Kodepogu et al., 2023).
Zlatkovic, M., Cvijovic, Z., Stevanovic, A., & Song, Y. (2023). Concepts of Signal Control Preemption for Emergency Vehicles in Connected Vehicle Environments. Journal of Road and Traffic Engineering, 69(2), 1-7. https://doi.org/10.31075/PIS.69.02.01
Sheela, S., Nataraj, K. R., & Mallikarjunaswamy, S. (2023). A Comprehensive Exploration of Resource Allocation Strategies within Vehicle Ad-Hoc Networks. Mechatron. Intell Transp. Syst., 2(3), 169-190. https://doi.org/10.56578/mits020305
Kan, H. Y., Li, C., & Wang, Z. Q. (2024). An Integrated Convolutional Neural Network-Bidirectional Long Short-Term Memory-Attention Mechanism Model for Enhanced Highway Traffic Flow Prediction. J. Urban Dev. Manag., 3(1), 18-33. https://doi.org/10.56578/judm030102
Trifunović, A., Senić, A., Čičević, S., Ivanišević, T., Vukšić, V., & Simović, S. (2024). Evaluating the Road Environment Through the Lens of Professional Drivers: A Traffic Safety Perspective. Mechatron. Intell Transp. Syst., 3(1), 31-38. https://doi.org/10.56578/mits030103
Stević, Ž., Subotić, M., Softić, E., & Božić, B. (2022). Multi-Criteria Decision-Making Model for Evaluating Safety of Road Sections. J. Intell. Manag. Decis., 1(2), 78-87. https://doi.org/10.56578/jimd010201
Kodepogu, K., Manjeti, V. B., & Siriki, A. B. (2023). Machine Learning for Road Accident Severity Prediction. Mechatron. Intell Transp. Syst., 2(4), 211-226. https://doi.org/10.56578/mits020403
Simić N, Ivanišević N, Nedeljković Đ, Senić A, Stojadinović Z, Ivanović M. Early Highway Construction Cost Estimation: Selection of Key Cost Drivers. Sustainability. 2023; 15(6):5584. https://doi.org/10.3390/su15065584
These references are my suggestions, you can use more references than this suggested to enrich the introduction with similar research in the world.
Response: We are grateful to the reviewers for providing us with so many references, which were very helpful in improving our introduction. We have expanded the introduction by including additional recent references to provide a broader perspective on the topic. These changes are mainly in paragraphs 2, 3, and 4 of the introduction, which are marked in red.
Comments 2: What is the accuracy rate of the vehicle recognition and collision detection components in diverse environmental conditions?
Response: Your suggestion to explore the accuracy in different environments is very meaningful. In order to verify the recognition accuracy of our trained yolov8 model in different environments, in Section 4.1, we tested the trained model again on the UA-DETRAC dataset. The experiment proved that yolov8 can effectively identify vehicles under four weather conditions: cloudy, night, rainy, and sunny. For the collision detection part, the data we used came from the Internet, which already included test results in various weather conditions. In addition, we did not find a public car accident dataset with environmental classification labels, or data or reference materials suitable for comparative experiments. Therefore, we did not add tests in different environments. However, we emphasized our analysis of collision detection results in the conclusion. In the results of this article, the high recall rate means that our model will not miss traffic accidents, and it is very effective as a low-computational screening detection model. We discussed the issue of data limitations in more detail in the newly added Section 5 discussion, including the limitations of this article at the data level and future work directions in data acquisition.
Comments 3:
What metrics were used to evaluate the accuracy and efficiency of the proposed method? How do these metrics compare to existing methods?
Response: In the vehicle recognition stage, we mainly evaluate four metrics: Recall, Precision, F1-Score and AP (average precision), and Figure 5 (line 345) shows the results of these metrics. In the tracking stage, the main comparison metric is “FPS”. In Table 1, we cite the literature for comparison. In the improved optical flow algorithm, we still use AEE, Angle error and FPS metrics to compare with the classic algorithm, and modify the visualization results to better show the advantages of the improved algorithm, this change can be found in Figure 9. To more effectively test the responsibility determinations generated by the model, we added more text information to test the accuracy of the determinations generated by the model in different scenarios. We also tested the reliability of the model’s determination results when key frames were missing or occluded. (see page 20，4.6.2. Experiment 2). We compared the information generated by the model with the actual determination results reported. Since there are few relevant literatures on automatic accident responsibility determination in the field of transportation, and even fewer articles on using large language models for responsibility determination, it is difficult to compare the accident responsibility determination scheme proposed in this paper with the schemes of existing articles.
Comments 4:
Are there any known limitations or potential challenges in implementing this method in real-world traffic systems?
Response: In response to this suggestion, we have added a special discussion section 5. We mainly discuss some existing technical issues, such as the hallucinations of large models and the improvement of the algorithms, as well as the problem of deploying the solutions proposed in this article in reality, and explore the solutions proposed in this article from a legal perspective.(see section 5 discussion)
Comments 5:
How does the system handle cases with incomplete or ambiguous data, such as missing frames or occluded vehicles?
Response: We have added Experiment 2 in Section 4.6.2. In Experiment 2, we tested the cases of missing key frames and vehicle information being blocked, and analyzed the experimental results.
Comments 5: The conclusion is missing future research directions, the limitations of the study, and the practical application of the obtained results.
Response: We have included a discussion of future work in Section 5, see 5.4 Future Work for details, discussing how to optimize the model in the future, how to deploy it in the real world, and the importance of obtaining more diverse datasets.
We believe that these revisions have significantly improved the quality and clarity of our manuscript. We hope that the changes address your concerns and enhance the overall contribution of our work.
Thank you once again for your valuable feedback.
Sincerely,
Junbo Chen
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsIn this manuscript, the authors propose a fully intelligent method for determining responsibility in minor accidents based on collision detection and large language models. However, I will comment on some aspects to improve the quality of the article, and the suggested changes should be highlighted:
- The authors must enrich and increase the Introduction Section
- The authors must add the Related Works Section.
- A brief introduction must appear after the title of a Section or Subsection.
- Was Figure 1 created by the authors? If this is not the case, it is necessary to include their bibliographical citation.
- In the title of 2.1.2, the authors must correctly place the double quotation marks.
- All acronyms must have their corresponding meaning within the text of the article the first time they appear.
- What is the reason why you have used YOLOv8? It would be best if you justified it within the article.
- In Line 271, check the correct way to write that sentence.
- The font size of the Figures must be more significant since they cannot be seen in detail.
- The object type (Table, Figure, Algorithm, etc.) after line 353 is not disclosed.
- The title of Section 5 must only be Conclusions.
- The authors must improve the conclusions.
- Correct Section 6.
Comments on the Quality of English LanguageIn this manuscript, the authors propose a fully intelligent method for determining responsibility in minor accidents based on collision detection and large language models. However, I will comment on some aspects to improve the quality of the article, and the suggested changes should be highlighted:
- The authors must enrich and increase the Introduction Section
- The authors must add the Related Works Section.
- A brief introduction must appear after the title of a Section or Subsection.
- Was Figure 1 created by the authors? If this is not the case, it is necessary to include their bibliographical citation.
- In the title of 2.1.2, the authors must correctly place the double quotation marks.
- All acronyms must have their corresponding meaning within the text of the article the first time they appear.
- What is the reason why you have used YOLOv8? It would be best if you justified it within the article.
- In Line 271, check the correct way to write that sentence.
- The font size of the Figures must be more significant since they cannot be seen in detail.
- The object type (Table, Figure, Algorithm, etc.) after line 353 is not disclosed.
- The title of Section 5 must only be Conclusions.
- The authors must improve the conclusions.
- Correct Section 6.
Author Response
Dear Reviewer,
Thank you very much for your thorough review and constructive feedback on our manuscript. We appreciate your insightful comments and suggestions, which have greatly contributed to improving the quality and clarity of our work. In the revised manuscript, the red font is the newly added content, and the highlighted part is the modified part. Below, we provide detailed responses to each of your comments and outline the specific revisions we have made to the manuscript.
Comments 1: The authors must enrich and increase the Introduction Section.
Response: We have expanded the Introduction Section to provide a more comprehensive background of the problem, the significance of the study, and the current state of research in this area. The revised Introduction now includes additional references and a more detailed explanation of the research context. These changes are mainly in paragraphs 2, 3, and 4 of the introduction, which are marked in red.
Comments 2: The authors must add the Related Works Section.
Response: We added some cutting-edge research in related fields to the introduction and introduced the research progress of relevant scholars. The changes are in the second and third paragraphs of the introduction.
Comments 3: A brief introduction must appear after the title of a Section or Subsection.
Response: We have revised each section and subsection to include a brief introductory paragraph that outlines the content and purpose of the following text. These introductions can be seen at the beginning of each section.
Comments 4: Was Figure 1 created by the authors? If this is not the case, it is necessary to include their bibliographical citation.
Response: Figure 1 was created by the authors. We still give a relevant reference in the text, which can be seen in line 100.
Comments 5: In the title of 2.1.2, the authors must correctly place the double quotation marks.
Response: We have corrected the placement of the double quotation marks. The revised title can be found on Section 2.2., line 105.
Comments 6: All acronyms must have their corresponding meaning within the text of the article the first time they appear.
Response: We have ensured that all acronyms are fully defined upon their first appearance in the manuscript. For example, "YOLOv8" is now defined in full on its first mention in line 84.
Comments 7: What is the reason why you have used YOLOv8? It would be best if you justified it within the article.
Response: We have added a justification for using YOLOv8, highlighting its advantages in terms of accuracy and efficiency for real-time vehicle detection. This justification is provided line 94.
Comments 8: check the correct way to write that sentence.
Response: We have reviewed and corrected the sentence on line 271 to improve clarity and grammatical accuracy. The revised sentence is “To assess tracking performance in real time, we recorded the PSR (Peak Sidelobe Ratio) for each frame. “, which can be found in line 372-373.
Comments 9: The font size of the Figures must be more significant since they cannot be seen in detail.
Response: We have increased the font size in all figures to ensure that the details are clearly visible. This adjustment has been applied to Figures 1, Figures 2, Figures 3, Figures 10, as seen throughout the manuscript.
Comments 10: The object type (Table, Figure, Algorithm, etc.) after line 353 is not disclosed.
Response: We confirmed the object type here in the original manuscript, and can be seen in line 567-568 of the revised manuscript. The revised version is “Table 7. The impact of additional textual information with subjective information on responsibility judgment.”
Comments 11: The title of Section 5 must only be Conclusions.
Response: We changed the title of Section 5 in the original manuscript to “6 Conclusions” in the revised manuscript according to your suggestion. A separate discussion of limitations and future work is added in Section 5 Discussion.
Comments 12: The authors must improve the conclusions.
Response: We have revised the Conclusions section to provide a more concise and impactful summary of our findings.
Comments 13: Correct Section 6.
Response: The original manuscript was reviewed and corrected in section 6, and the revised version is in section 7
We hope these revisions address your concerns and improve the overall quality of our manuscript. Thank you again for your valuable feedback.
Best regards,
Junbo Chen
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsThe article is well done and significant. The authors proposed a new approach to determining responsibility in minor accidents based on modern AI methods. The article's theme has a practical value for eliminating human parts to estimate the responsibility in minor accidents and helps to do the estimation completely digitally. The advance of the article is the modern approach based on the YOLOv8 algorithm, combined with a minimum mean square error filter for real-time target tracking. The disadvantage is the small introduction and state-of-art part.
There are such recommendation
1. It would be better to briefly describe the state of the art.
2. In Figure 8 it is hard to understand the difference between pictures A) and B). Maybe It would be better to present only the important part of the picture or choose the other way for visualization.
3. It would be better to add the math symbol "min" in the right part of the formula (17).
4. The authors described x and y as coordinate directions in the image (see lines 97). It is unclear if x and y are the same in the notation flow vector (u_{x,y,t};v_{x,y,t}) (line 105). It would be better to detail this and also add the description index t for the flow vector.
5. The authors used index I for the notation pixel of the image (see line 151) and the number for the image (see line 160). It has to be the different notations.
6. In line 154 the authors write "in the window (𝑖,𝑗)". it is a mistake because (I,j) is not a window but the pixel of an image.
7. Formula (15) is also used index t (see notice 4). Maybe it would be better to use other notation for summing?
8. Table 4 is hard for the susceptibility. Would be better to use a small font? or divide Table 4 into two tables?
9. It would be better to add more plots with numerical experiments
(like Figures 4 and 5). Because using the modern approaches for simulation can be easily illustrated by plots.
After minor revision, the article can be published
Author Response
Dear Reviewer
We sincerely appreciate the time and effort you have dedicated to reviewing our manuscript. Your detailed and constructive comments have been instrumental in enhancing the clarity, accuracy, and depth of our study. We are grateful for the opportunity to refine our work based on your insightful feedback, which has undoubtedly improved the quality of our paper. In the revised manuscript, the red font is the newly added content, and the highlighted part is the modified part.
Comments 1: It would be better to briefly describe the state of the art.
Response: We have expanded the introduction sections to provide a more comprehensive overview of existing methods and their limitations in automated responsibility determination for minor traffic accidents. These changes are mainly in paragraphs 2, 3, and 4 of the introduction, which are marked in red
Comments 2: In Figure 8 it is hard to understand the difference between pictures A) and B). Maybe it would be better to present only the important part of the picture or choose the other way for visualization.
Response: To clarify the differences between images A) and B) in Figure 7 (now Figure 9 in revised version), we have zoomed in on the key areas where our improvements are most evident. Additionally, we added descriptive annotations directly onto the images to guide the reader through the specific enhancements achieved with our methodology. (see line 400-404)
Comments 3: It would be better to add the math symbol "min" in the right part of the formula (17).
Response: We have revised Formula (17) to include the 'min' symbol explicitly, ensuring the mathematical expression aligns correctly with the standard optimization notation. This change is highlighted and can be found in Formula (21).
Comments 4: The authors described x and y as coordinate directions in the image (see lines 97). It is unclear if x and y are the same in the notation flow vector (u_{x,y,t};v_{x,y,t}) (line 105). It would be better to detail this and also add the description index t for the flow vector.
Response: We have clarified the notation in the description of the flow vectors to ensure consistency throughout the manuscript. We now explicitly define and as spatial coordinates and as the temporal coordinate in both the text and the mathematical expressions, eliminating any ambiguity.
Comments 5: The authors used index I for the notation pixel of the image (see line 151) and the number for the image (see line 160). It has to be the different notations.
Response: To clarify and correct this notation, we have adjusted the manuscript to ensure distinct and consistent symbols are used throughout. The pixel notation has been changed. (see line 211). This revision eliminates any potential confusion between image pixels and image indices, improving the readability and technical accuracy of our manuscript.
Comments 6: In line 154 the authors write "in the window (?,?)". it is a mistake because (I,j) is not a window but the pixel of an image.
Response: We thank the reviewer for pointing out this error. In the modified version, it becomes “ at the pixel location ”. This change ensures that the terminology aligns with standard image processing language.
Comments 7: Formula (15) is also used index t (see notice 4). Maybe it would be better to use other notation for summing?
Response:
We have revised the indexing in Formula (15) to avoid overlap with the temporal index . The summation indices have been adjusted to “” to ensure clarity and prevent any notational confusion with the time index. (see formula 16)
Comments 8: Table 4 is hard for the susceptibility. Would be better to use a small font? or divide Table 4 into two tables?
Response:
Font sizes and table formatting have been adjusted to ensure the data is accessible and easy to interpret. The modified version is in Table 7.
Comments 9: It would be better to add more plots with numerical experiments (like Figures 4 and 5). Because using the modern approaches for simulation can be easily illustrated by plots.
Response:
We have included additional plots to illustrate the results of our numerical experiments more comprehensively. These new figures showcase the robustness and efficacy of our approach under various conditions, providing a clearer visualization of the performance improvements. Figure 5 and Figure 13 are the additional plots.
We hope these revisions address your concerns and improve the overall quality of our manuscript. Thank you again for your valuable feedback.
Best regards,
Junbo Chen
Author Response File: Author Response.pdf
Reviewer 4 Report
Comments and Suggestions for AuthorsThe manuscript describes research conducted to develop a fully intelligent method for determining responsibility in minor traffic accidents without human intervention. The issue addressed under the approach proposed is new and interesting.
I missed an interesting reference on the same issue in the literature review:
Liu, S., Zhang, Z. J., & Yu, Z. H. (2022). Research on liability Identification System of Road Traffic Accident. Journal of Computers, 33(1), 215-224.
In general, the methods are well-described. However, the validation of the research would benefit from more detailed descriptions of the experimental setup, including the specific parameters used in the YOLOv8 training, the dataset details, and the conditions under which the experiments were conducted.
Since the method aims to reduce manual intervention, it would be helpful to include a section on how users (e.g., traffic authority) interact with the system, including any user interface or feedback mechanisms.
Figure 8 needs further explanation
Figure 9. Do the authors believe that the metrics obtained are good enough so as to settle legal matters of these nature. Please, justify.
Section 4.3. Is there a traffic authority report on the case presented with which to compare the report generated by the tool?
Without detracting from the work of the researchers, I consider that in any case these tools, until further technical and regulatory advances are made, should serve at most to assist the traffic authorities and eventually in court cases to help the magistrates in their decision making.
It has been found that the most advanced AI-based language models still suffer from hallucinations when they have to refer to external references, such as the laws and regulations cited in the document. This is a possible limitation that should be taken into account.
McIntosh, T. R., Liu, T., Susnjak, T., Watters, P., Ng, A., & Halgamuge, M. N. (2023). A culturally sensitive test to evaluate nuanced gpt hallucination. IEEE Transactions on Artificial Intelligence.
Lee, M. (2023). A mathematical investigation of hallucination and creativity in GPT models. Mathematics, 11(10), 2320.
Sovrano, F., Ashley, K., & Bacchelli, A. (2023, July). Toward eliminating hallucinations: Gpt-based explanatory ai for intelligent textbooks and documentation. In CEUR Workshop Proceedings (No. 3444, pp. 54-65). CEUR-WS.
The research presented in the document raises several ethical concerns related to the use of AI and advanced technologies in determining liability for minor traffic accidents. The use of vehicle recognition and real-time tracking involves extensive surveillance, which can lead to privacy violations if not properly managed. AI models can inherit biases from the data they are trained on. The use of AI in legal contexts, such as liability determination, raises questions about the legal status of AI decisions and the rights of individuals to contest these decisions. Moreover, the decision-making process of AI systems should be transparent to ensure that the outcomes are understandable and justifiable to all parties involved. I suggest the authors to, at least, discuss these issues in their manuscript.
Please include the above discussion and limitations in the section “Conclusion and outlook”.
I believe that the images with examples of accidents clearly require informed consent statement of the people involved.
Comments on the Quality of English LanguageSome minor English mistakes and typos are found in the document.
Author Response
Dear Reviewer,
Thank you very much for your thorough review and constructive feedback on our manuscript. We appreciate your insightful comments and suggestions, which have greatly contributed to improving the quality and clarity of our work. The red font is the new content, the highlighted part is the change that needs attention, and the whole article has been polished. Below, we provide detailed responses to each of your comments and outline the specific revisions we have made to the manuscript.
Comments 1: The manuscript describes research conducted to develop a fully intelligent method for determining responsibility in minor traffic accidents without human intervention. The issue addressed under the approach proposed is new and interesting.
I missed an interesting reference on the same issue in the literature review:
Liu, S., Zhang, Z. J., & Yu, Z. H. (2022). Research on Liability Identification System of Road Traffic Accident. Journal of Computers, 33(1), 215-224.
Response: Thank you for your positive feedback on the novelty of our approach. We have added the suggested reference to our literature review to provide a more comprehensive background on the topic. The reference can be found in the revised manuscript in the literature review section, line 66. We have also added more cutting-edge literature and enriched the introduction section.
Comments 2: In general, the methods are well-described. However, the validation of the research would benefit from more detailed descriptions of the experimental setup, including the specific parameters used in the YOLOv8 training, the dataset details, and the conditions under which the experiments were conducted.
Response: We appreciate your suggestion to include more details on the experimental setup. We have expanded the methods section to describe the specific parameters used in the YOLOv8 training, the dataset details, and the experimental conditions. In order to clearly describe the parameters, we created Table 3, which shows the parameters for training the yolov8 network. The other methods and experimental parts in the article have also been modified, adding more descriptions and evaluations
Comments 3: Since the method aims to reduce manual intervention, it would be helpful to include a section on how users (e.g., traffic authority) interact with the system, including any user interface or feedback mechanisms.
Response: Thank you for this valuable suggestion. This suggestion is invaluable when engineering the methods proposed in this paper. However, considering the overall scope of the article, we did not find a particularly suitable place to add a new section. We discuss how users interact with the system in Section 5 Discussion, future progress related to user interaction is added.
Comments 4: Figure 8 needs further explanation.
Response: We have revised the caption of Figure 8 to provide a more detailed explanation of the figure, including what each component represents and how it contributes to the overall analysis. Figure 9 in the revised manuscript is Figure 8 in the original manuscript, and the added explanation is in line 400-404.
Comments 5: Figure 9. Do the authors believe that the metrics obtained are good enough so as to settle legal matters of this nature? Please, justify.
Response: We have included a discussion in the revised manuscript addressing the sufficiency of the metrics obtained in Figure 9 (Figure 11 in revised manuscript) for settling legal matters. We discuss the reliability and robustness of the metrics in relation to current legal standards, acknowledging the limitations. The explanation in line 479-484 mentions the recall value of 1, which shows that the model is very effective as a filter and will not miss traffic accidents, which is also of great significance for traffic regulation.
Comments 6: Section 4.3. Is there a traffic authority report on the case presented with which to compare the report generated by the tool?
Response: Yes, there is a traffic authority report available for the case presented. Reference 35 gives the URL of the report. We have now included a comparison between the traffic authority’s report and the report generated by our tool.
Comments 7: Without detracting from the work of the researchers, I consider that in any case these tools, until further technical and regulatory advances are made, should serve at most to assist the traffic authorities and eventually in court cases to help the magistrates in their decision-making.
Response: The solution we have proposed at present still has a long way to go to completely replace humans. However, the accidents targeted in this article are minor traffic accidents, not major traffic accidents. Because such accidents often do not involve casualties, the purpose of rapid traffic evacuation can be achieved by simply communicating and coordinating with on-site personnel. The solution proposed in this article can be developed into a mobile app (in fact, we are working on it). On-site personnel can use such a solution to quickly determine responsibility without waiting for the traffic police to arrive. Therefore, once the on-site personnel agree on the responsibility determination, the traffic can be evacuated. If the result of the determination of responsibility is not enough to convince the on-site personnel, they can still wait for the traffic police to come and determine the responsibility. Therefore, for now, the solution for handling traffic accidents proposed in this article is still of great significance to traffic safety.
Reviewer Comment 8: It has been found that the most advanced AI-based language models still suffer from hallucinations when they have to refer to external references, such as the laws and regulations cited in the document. This is a possible limitation that should be taken into account.
McIntosh, T. R., Liu, T., Susnjak, T., Watters, P., Ng, A., & Halgamuge, M. N. (2023). A culturally sensitive test to evaluate nuanced GPT hallucination. IEEE Transactions on Artificial Intelligence.
Lee, M. (2023). A mathematical investigation of hallucination and creativity in GPT models. Mathematics, 11(10), 2320.
Sovrano, F., Ashley, K., & Bacchelli, A. (2023, July). Toward eliminating hallucinations: Gpt-based explanatory ai for intelligent textbooks and documentation. In CEUR Workshop Proceedings (No. 3444, pp. 54-65). CEUR-WS.
Response: Thank you for this suggestion. We have acknowledged the issue of AI model hallucinations in the context of legal references and regulations. A discussion of this limitation has been added to the “5.1 Limitations of the Algorithm “section and “5.4 Future Work”, where we also cited the suggested references to reinforce our argument. This addition can be found in line 626.
Comments 9: The research presented in the document raises several ethical concerns related to the use of AI and advanced technologies in determining liability for minor traffic accidents. The use of vehicle recognition and real-time tracking involves extensive surveillance, which can lead to privacy violations if not properly managed. AI models can inherit biases from the data they are trained on. The use of AI in legal contexts, such as liability determination, raises questions about the legal status of AI decisions and the rights of individuals to contest these decisions. Moreover, the decision-making process of AI systems should be transparent to ensure that the outcomes are understandable and justifiable to all parties involved. I suggest the authors to, at least, discuss these issues in their manuscript.
Please include the above discussion and limitations in the section “Conclusion and Outlook”.
Response: We appreciate your thoughtful comments on the ethical implications of our research. We have included a discussion of these ethical concerns in the “5.3 Legal and Ethical Considerations” section, Privacy issues, potential bias in AI models, the legal status of AI decisions, and the need for transparency in AI decision-making processes are discussed.
Comments 11: I believe that the images with examples of accidents clearly require informed consent statement of the people involved.
Response: We acknowledge the importance of informed consent for using images in our research. All accident images in this article are from public reports.
Comments 12: Comments on the Quality of English Language: Some minor English mistakes and typos are found in the document.
Response: We have thoroughly reviewed the manuscript to correct minor English mistakes and typos. The language has been refined for clarity and consistency throughout the document.
We hope these revisions address your concerns and improve the overall quality of our manuscript. Thank you again for your valuable feedback.
Best regards,
Junbo Chen
Author Response File: Author Response.pdf
Reviewer 5 Report
Comments and Suggestions for AuthorsThe purpose of this article is to propose and validate an autonomous method for determining liability in minor traffic accidents using collision detection and large language models. The method aims to eliminate human intervention, improving efficiency and accuracy in accident reporting. Simulation experiments demonstrate the feasibility and effectiveness of the proposed approach in real-time accident analysis and responsibility determination.
Based on a review of the article, here are specific points that need enhancement, improvement, or correction:
1)The abstract could benefit from a clearer structure. Specifically, delineate the background, methodology, results, and conclusions more distinctly. That would enhance readability and comprehension.
2)In subsection 2.1.1, the explanation of the YOLOv8 network design could be made clearer by including more detailed descriptions or visual aids for the non-specialist reader. Explain the role and benefits of the C2f (Coarse-to-Fine) module in simpler terms.
3)Equations throughout the text, such as in sections 2.1.2 and 3.2.3, need better explanations for the variables used. In Equation (1), clarify what each term represents directly after the equation.
4)In section 3.3.1, the improved Horn & Schunck algorithm is mentioned. More details on how this improvement is implemented and its advantages over the classic version would be helpful.
5)The methodology for key frame extraction (section 3.4.1) should include more visual examples to illustrate how K-means clustering is applied to video frames.
6)Section 4 could benefit from a more structured presentation of results. Use subheadings such as "Vehicle Recognition Accuracy," "Tracking Performance," and "Collision Detection Efficacy" to organize the findings.
7)Provide more context for the supplementary text information explored in section 4.3.2. Explain why these specific scenarios were chosen and how they contribute to validating the model.
8)Include more comparative data and discussion on how the proposed methods perform relative to existing solutions.
9)Figures such as Figure 1 (C2f Module Structure Diagram) and Figure 7 (Optical Flow Estimation Comparison Results) should have more descriptive captions.
10)The flowcharts in Figures 2 and 3 should be simplified for better readability. Use larger fonts and distinct colours to differentiate between the steps.
11)The conclusion section needs a clearer summary of the findings and their implications. Additionally, provides a more detailed outlook on future work, specifying the potential improvements and applications of the proposed method.
Comments on the Quality of English LanguageModerate editing of the English language is required.
Improve language consistency and correct grammatical errors throughout the manuscript. For instance, ensure that technical terms are consistently used and check for subject-verb agreement.
Author Response
Response to Reviewer
Dear Reviewer,
We would like to express our sincere gratitude for your thoughtful and constructive feedback on our manuscript titled "An Autonomous Intelligent Liability Determination Method for Minor Accidents Based on Collision Detection and Large Language Models." Your comments have been invaluable in improving the clarity, depth, and overall quality of our work. The red font is the new content, the highlighted part is the change that needs attention, and the whole article has been polished. Below, we provide a detailed response to each of your suggestions:
1) Abstract Structure Improvement:
Comment 1: The abstract could benefit from a clearer structure. Specifically, delineate the background, methodology, results, and conclusions more distinctly. That would enhance readability and comprehension.
Response: We have restructured the abstract to clearly delineate the background, methodology, results, and conclusions. This improves readability and provides a more structured summary of our research.
2) Clarification of YOLOv8 Network Design:
Comment 2: In subsection 2.1.1, the explanation of the YOLOv8 network design could be made clearer by including more detailed descriptions or visual aids for the non-specialist reader. Explain the role and benefits of the C2f (Coarse-to-Fine) module in simpler terms.
Response: The explanation of the YOLOv8 network design has been made clearer. We included more detailed descriptions and a visual aid to make the content accessible to non-specialist readers. The role and benefits of the C2f module are now explained in simpler terms. This change can be found in line 88-94.
3) Explanation of Variables in Equations:
Comment: Equations throughout the text, such as in sections 2.1.2 and 3.2.3, need better explanations for the variables used. In Equation (1), clarify what each term represents directly after the equation.
Response: We added explanations for the variables used in key equations, particularly in sections 2.1.2 and 3.2.3. Each term in Equation (1) is now clearly defined immediately after the equation. We ensured that variables were consistent and that no variable had multiple meanings.
4) Details on Improved Horn & Schunck Algorithm:
Comment: In section 3.3.1, the improved Horn & Schunck algorithm is mentioned. More details on how this improvement is implemented and its advantages over the classic version would be helpful.
Response: We have expanded section 3.3.1 to include a detailed explanation of the improvements made to the Horn & Schunck algorithm. We also discuss the advantages of the improved algorithm compared to the classic version, highlighting its enhanced performance in collision detection scenarios. This change can be found highlighted in line 241-245. In Section 4.4.1., we also emphasize the superiority of the improved algorithm in Figure 9 and describe it in detail.
5) Visual Examples for Key Frame Extraction:
Comment: The methodology for key frame extraction (section 3.4.1) should include more visual examples to illustrate how K-means clustering is applied to video frames.
Response: In section 3.4.1, we included visual examples to illustrate the application of K-means clustering to video frames, aiding the reader's understanding of the methodology. We realized that it was not clear how the algorithm extracted features, so we added Figure 3 and some mathematical formulas to describe it. This was modified on page 7. In the experimental results presentation, we have separately listed Section 4.5.1 as a presentation of the K-means algorithm clustering, and added Figures 12 and 13 to more clearly describe the clustering process.
6) Structured Presentation of Results:
Comment: Section 4 could benefit from a more structured presentation of results. Use subheadings such as "Vehicle Recognition Accuracy," "Tracking Performance," and "Collision Detection Efficacy" to organize the findings.
Response: Section 4 has been restructured with subheadings to better organize the presentation of results. This restructuring aims to improve the readability and clarity of the findings. Subheadings include: “4.1 Experimental parameter setting”, “4.2 Vehicle Recognition Accuracy”, “4.3 Tracking Performance”, “4.4 Collision Detection Efficacy”, “4.5 Accident Liability Assessment” “4.6 Analysis of the Impact of Image and Text Information on Liability Determination Results”
7) Context for Supplementary Text Information:
Comment: Provide more context for the supplementary text information explored in section 4.3.2. Explain why these specific scenarios were chosen and how they contribute to validating the model.
Response: We have revised section 4.3.2 to provide additional context for the supplementary text information used in our experiments. We explain the rationale behind selecting these specific scenarios and how they help in validating the proposed model. We also conducted a more in-depth analysis of these text information and judgment results. These changes can be found in table 7. We also realized that incomplete image information also affects the results, and added Section 4.6.2 to explore this issue.
8) Comparative Data and Discussion:
Comment: Include more comparative data and discussion on how the proposed methods perform relative to existing solutions.
Response: In terms of data, we added tests using the UA-DETRAC dataset in Section 4.2 Vehicle Recognition Accuracy to explore the recognition performance of the model under different weather conditions. In Section 4.6, in order to more effectively test the liability determination results generated by the model, we added more text information to test the accuracy of the model's determination results in different scenarios (see page 17, 4.6.2 Experiment 1), and tested the reliability of the model's determination results when key frames are missing or occluded (see page 20, 4.6.2 Experiment 2). We compared the information generated by the model with the actual reported determination results.
However, since there are few relevant literatures on automatic accident liability determination in the field of transportation, and even fewer articles using large language models for liability determination, it is difficult to compare the performance of the accident liability determination scheme proposed in this article with the schemes of existing articles.
9) Descriptive Captions for Figures:
Comment: Figures such as Figure 1 (C2f Module Structure Diagram) and Figure 7 (Optical Flow Estimation Comparison Results) should have more descriptive captions.
Response: We have revised the captions for Figures 1 and 7 to be more descriptive, providing better explanations of what the figures represent and their significance in the context of our study. The title of Figure 1 has been changed to "C2f Module Structure Diagram. The core component of the C2f module is the ConBNSiLU module, which combines convolution (Conv), batch normalization (BN), and SiLU activation function." can be seen in line 102. The original Figure 7 is Figure 10 in the revised version, and is renamed "Visualization comparison results of optical flow estimation (a) Original Horn & Schunck algorithm optical flow calculation results; (b) Improved Horn & Schunck algorithm optical flow calculation results." can be seen in line 397.
10) Simplification of Flowcharts:
Comment: The flowcharts in Figures 2 and 3 should be simplified for better readability. Use larger fonts and distinct colors to differentiate between the steps.
Response: We have simplified the flowcharts in Figures 2 and 3 by increasing font sizes and using distinct colors to clearly differentiate between the steps. This should enhance the readability and visual appeal of the flowcharts.
11) Clearer Conclusion and Future Work:
Comment: The conclusion section needs a clearer summary of the findings and their implications. Additionally, provide a more detailed outlook on future work, specifying the potential improvements and applications of the proposed method.
Response: We added a discussion section before the conclusion, which mainly discusses the limitations of the algorithm proposed in this paper, its application potential, legal issues, and future improvements. We put the conclusion in Section 6 and rewrote it to make it more logical.
We believe that these revisions have significantly improved the quality and clarity of our manuscript. We hope that the changes address your concerns and enhance the overall contribution of our work.
Thank you once again for your valuable feedback.
Sincerely,
Junbo Chen
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report
Comments and Suggestions for AuthorsThank you to the authors for making the suggested changes. A minor revision to the manuscript written in English is required before publication.
Comments on the Quality of English LanguageThank you to the authors for making the suggested changes. A minor revision to the manuscript written in English is required before publication.
Author Response
Dear Reviewer,
Thank you for your detailed review and the insights you have provided. We have carefully considered each comment and have made appropriate revisions to the manuscript to address your concerns. Below, we detail our responses to your comments and the actions we have taken.
Comments 1: Thank you to the authors for making the suggested changes. A minor revision to the manuscript written in English is required before publication.
Response: Thank you for your feedback and the encouragement to refine our manuscript further. We appreciate your recognition of the efforts made to address previous comments.
Following your suggestion for a minor revision, we have thoroughly reviewed the manuscript to enhance its readability and ensure clarity in the English language. We have engaged a native English-speaking editor to ensure that the text meets academic standards. The revisions span the entire document, with particular attention to areas previously noted for improvement.
We have also made some additional modifications. We revised the captions of most images to make them more understandable. We removed irrelevant references to make the introduction more relevant to the topic. Moreover, we provided a more detailed description of the datasets used, emphasizing their public availability. We believe these modifications have improved the quality of the manuscript. We hope that these changes will be approved for publication.
Best regards,
Junbo Chen
Author Response File: Author Response.pdf
Reviewer 4 Report
Comments and Suggestions for AuthorsThe authors have addressed some of my comments. However, there are still some issues that need to be resolved before the manuscript is considered worthy of publication.
Comment #1:
There are contradictions in the text that need correction that are related to my previous comment regarding ethical concerns. I consider that in any case these tools, until further technical and regulatory advances are made, should serve at most to assist the traffic authorities and eventually in court cases to help the magistrates in their decision-making.
Line 75. This method aims to eliminate human intervention entirely, thereby improving the efficiency and objectivity of accident reporting.
Line 275. The authors claim that their model can comprehensively consider information from multiple data sources to achieve accurate liability determination and reduce the need for manual intervention.
Comment #2:
The new pieces of text inserted as responses to reviewer comments have poor English language. For example, lines 401-405.
Comment #3:
I have detected new references inserted in the manuscript that do not seem to align with the content and context of the study. Upon further examination, these citations appear to be unrelated to the research topic and lack relevance to the research work conducted by the authors. I suspect that other reviewers might have suggested citations intended to artificially inflate their citation count of certain publications rather than to enhance the quality of the manuscript.
5. Simić N, Ivanišević N, Nedeljković Đ, Senić A, Stojadinović Z, Ivanović M. Early Highway Construction Cost Estimation: Se-709 lection of Key Cost Drivers. Sustainability. 2023; 15(6):5584. https://doi.org/10.3390/su15065584.
6. Trifunović, A., Senić, A., Čičević, S., Ivanišević, T., Vukšić, V., & Simović, S. (2024). Evaluating the Road Environment Through 711 the Lens of Professional Drivers: A Traffic Safety Perspective. Mechatron. Intell Transp. Syst., 3(1), 31-38. https://doi.org/10.56578/mits030103.
8. Zlatkovic, M., Cvijovic, Z., Stevanovic, A., & Song, Y. (2023). Concepts of Signal Control Preemption for Emergency Vehicles in 717 Connected Vehicle Environments. Journal of Road and Traffic Engineering, 69(2), 1-7. https://doi.org/10.31075/PIS.69.02.01.
10. Stević, Ž., Subotić, M., Softić, E., & Božić, B. (2022). Multi-Criteria Decision-Making Model for Evaluating Safety of Road Sec-721 tions. J. Intell. Manag. Decis., 1(2), 78-87. https://doi.org/10.56578/jimd010201.
The introduction in the original version of the manuscript was fine, well suited to the topic. It was not necessary to add all these non-relevant references.
Comment #4:
The authors have not indicated the necessary consent of the people involved in the accidents used as examples. To be indicated as appropriate under the text “Institutional Review Board Statement” and “Informed Consent Statement”. “Not applicable” is not the appropriate selection.
Comment #5:
Most figure captions are still meagre. Further descriptions are needed as well as the sources for those that were not taken by the authors.
Comments on the Quality of English LanguageThe new pieces of text inserted as responses to reviewer comments have poor English language. For example, lines 401-405.
Author Response
Dear Reviewer,
Thank you for your detailed review and the insights you have provided. We have carefully considered each comment and have made appropriate revisions to the manuscript to address your concerns. Below, we detail our responses to your comments and the actions we have taken.
Comments 1: There are contradictions in the text that need correction that are related to my previous comment regarding ethical concerns. I consider that in any case these tools, until further technical and regulatory advances are made, should serve at most to assist the traffic authorities and eventually in court cases to help the magistrates in their decision-making.
Line 75. This method aims to eliminate human intervention entirely, thereby improving the efficiency and objectivity of accident reporting.
Line 275. The authors claim that their model can comprehensively consider information from multiple data sources to achieve accurate liability determination and reduce the need for manual intervention.
Response: Thank you for highlighting the contradictions related to our system’s capabilities and ethical concerns in your comments. We have carefully revised the manuscript to address these issues, specifically in lines 63-69 and 259-262. In these sections, we re-emphasize that our system is primarily designed as a fully autonomous solution for on-site accident handling.
After thorough consideration, we decided to maintain the positioning of our system as an autonomous solution rather than just an assistive tool. We believe that defining it as such not only reflects our innovative approach but also adds value to its practical applications. This perspective fosters a direction for our ongoing efforts that is both ambitious and aligned with the future advancements in technology and regulations.
We acknowledge, as you rightly pointed out, that due to current technological and regulatory limitations, our system cannot yet fully replace human intervention. We have discussed these limitations extensively in the discussion section of our manuscript, particularly in lines 604-617. Here, we outline potential applications of our system within the current framework. For instance, it can assist traffic authorities by providing preliminary liability determinations at accident scenes or act as an initial screener to resolve disputes before escalating to human officers, significantly conserving police resources. Additionally, as you suggested, it can serve as a supportive tool in legal settings to aid magistrates in their decision-making.
We hope that these revisions and clarifications meet your satisfaction. We strive to balance innovation with practical application, always considering the current limitations while preparing for future advancements. Thank you for your valuable feedback, which has been instrumental in refining our work and ensuring its relevance and rigor.
Comments 2: The new pieces of text inserted as responses to reviewer comments have poor English language. For example, lines 401-405.
Response: We apologize for the oversight in the language quality of the newly inserted text. We have now subjected the entire manuscript, including these sections, to a thorough review by a native English-speaking editor to ensure clarity and grammatical accuracy. These revisions have been implemented in lines 401-405 and throughout the document.
Comments 3: I have detected new references inserted in the manuscript that do not seem to align with the content and context of the study. Upon further examination, these citations appear to be unrelated to the research topic and lack relevance to the research work conducted by the authors. I suspect that other reviewers might have suggested citations intended to artificially inflate their citation count of certain publications rather than to enhance the quality of the manuscript.
- Simić N, Ivanišević N, Nedeljković Đ, Senić A, Stojadinović Z, Ivanović M. Early Highway Construction Cost Estimation: Se-709 lection of Key Cost Drivers. Sustainability. 2023; 15(6):5584. https://doi.org/10.3390/su15065584.
- Trifunović, A., Senić, A., Čičević, S., Ivanišević, T., Vukšić, V., & Simović, S. (2024). Evaluating the Road Environment Through 711 the Lens of Professional Drivers: A Traffic Safety Perspective. Mechatron. Intell Transp. Syst., 3(1), 31-38. https://doi.org/10.56578/mits030103.
- Zlatkovic, M., Cvijovic, Z., Stevanovic, A., & Song, Y. (2023). Concepts of Signal Control Preemption for Emergency Vehicles in 717 Connected Vehicle Environments. Journal of Road and Traffic Engineering, 69(2), 1-7. https://doi.org/10.31075/PIS.69.02.01.
- Stević, Ž., Subotić, M., Softić, E., & Božić, B. (2022). Multi-Criteria Decision-Making Model for Evaluating Safety of Road Sec-721 tions. J. Intell. Manag. Decis., 1(2), 78-87. https://doi.org/10.56578/jimd010201.
The introduction in the original version of the manuscript was fine, well suited to the topic. It was not necessary to add all these non-relevant references.
Response: Upon reviewing the references you mentioned, we agree that they were not directly relevant to our study's focus and have removed them. We have ensured that all citations directly support the research and are pertinent to our study’s framework. This correction has streamlined the reference list and reinforced the manuscript's focus.
Comments 4: The authors have not indicated the necessary consent of the people involved in the accidents used as examples. To be indicated as appropriate under the text “Institutional Review Board Statement” and “Informed Consent Statement”. “Not applicable” is not the appropriate selection.
Response: We have revised our Institutional Review Board Statement and Informed Consent sections to reflect the sourcing of publicly available data and the non-requirement for direct consent. We clarified that the data used, being publicly accessible and not involving personal identifiers, complies with ethical standards for research.
Comments 5: Most figure captions are still meagre. Further descriptions are needed as well as the sources for those that were not taken by the authors.
Response: We acknowledge that the original captions were insufficiently detailed. We have now expanded each figure caption to include a clear description of what is depicted and sourced correctly where necessary. This enhancement not only complies with academic standards but also provides readers with a better understanding of each figure’s relevance and context.
Once again, we thank you for your constructive comments, which have undoubtedly strengthened our manuscript. We appreciate your attention to detail and hope that our revisions meet your approval.
Best regards,
Junbo Chen
Author Response File: Author Response.pdf
Reviewer 5 Report
Comments and Suggestions for AuthorsVery good revision.
Accept in its present form
Author Response
Dear Reviewer,
Thank you very much for your approval and positive feedback on our revisions. We appreciate your guidance and support throughout the review process, which has undoubtedly helped us improve the quality of our work.
We have made some additional changes to make the article higher quality. We revised the captions of most images to make them more understandable. We removed irrelevant references to make the introduction more relevant to the topic. Moreover, we provided a more detailed description of the datasets used, emphasizing their public availability. We believe these modifications have improved the quality of the manuscript.
Thank you once again for the opportunity to contribute to applied sciences.
Best regards,
Junbo chen
Author Response File: Author Response.pdf
Round 3
Reviewer 4 Report
Comments and Suggestions for AuthorsAll my major concerns have been addressed. Only minor language issues are pending.
Comments on the Quality of English LanguageMinor language and style issues should be corrected. For example
Line 100:
"David S. Bolme, J. Ross Beveridge, and others proposed the use of the MOSSE (Minimum Output Sum of Squared Error) filter [18] for target tracking."
Should read
"Bolme et al. [18] proposed the use of the MOSSE (Minimum Output Sum of Squared Error) filter for target tracking."
Line 229:
"Utilizing the HS algorithm, the energy functional of the optical flow vector, as described in Formula 4, is derived."
should read
The HS algorithm is used to derive the energy functional of the optical flow vector as described in Equation (4).
Author Response
Dear Reviewer,
Thank you for your thorough review and for acknowledging that the major concerns with our manuscript have been addressed. We appreciate your continued guidance and the additional language suggestions you have provided to enhance the clarity of our text.
We have carefully revised the manuscript to correct the minor language and style issues as per your recommendations:
- Line 99 Revised:
Previous: "David S. Bolme, J. Ross Beveridge, and others proposed the use of the MOSSE (Minimum Output Sum of Squared Error) filter [18] for target tracking."
Revised to: "Bolme et al. [18] proposed the use of the MOSSE (Minimum Output Sum of Squared Error) filter for target tracking." - Line 221 Revised:
Previous: "Utilizing the HS algorithm, the energy functional of the optical flow vector, as described in Formula 4, is derived."
Revised to: "The HS algorithm is used to derive the energy functional of the optical flow vector as described in Equation (4)."
We have carefully reviewed the entire article and made additional changes to address language and style issues.
- Line 128 Revised:
Previous: "The Horn & Schunck (HS) algorithm is a widely-utilized method for estimating optical flow vector fields [20][21][22], entailing the calculation of velocity and direction for each pixel across a sequence of images."
Revised to: "The HS (Horn & Schunck) algorithm, extensively used for estimating optical flow vector fields [20][21][22], computes velocity vectors that indicate each pixel's direction and speed between consecutive images in a sequence."
- Line 142 Revised:
Previous: “The process of estimating the optical flow vector field involves minimizing the aforementioned energy functional.”
Revised to: “Estimating the optical flow vector field involves minimizing the energy functional presented in Equation (4).”
- Line 192 Revised:
Previous: “To concentrate on the window's center, a cosine window function reduces the pixel values near the tracking window's edges.”
Revised to: “To focus on the center of the tracking window, a cosine window function is applied to taper the pixel values towards the edges.”
- Line 284 Revised:
Previous: “For each frame feature V_t , based on its distance to each cluster center C_k, it is assigned to the nearest cluster, updating the cluster assignment based on the smallest ‖V_t-C_k ‖^2.”
Revised to: “Each frame feature V_t is assigned to the nearest cluster by minimizing the squared Euclidean distance to the corresponding cluster center C_k.”
- Line 355 Revised:
Previous: “The PSR for each frame was calculated using this configuration:”
Revised to: “The PSR for each frame was calculated as follows:”
We have implemented these changes throughout the manuscript to ensure consistency and precision in our language use. We hope that these modifications meet your expectations and that the manuscript is now ready for publication.
Thank you once again for your valuable feedback and for helping us improve the quality of our work. Please let us know if there are any further adjustments we should make.
Best regards,
Junbo Chen
Author Response File: Author Response.pdf
