Temporal Margins and Behavioral Features for Early Risk Assessment in Left-Turn Vehicle and Bicycle Conflicts at Signalized Intersections

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This study analyzes collision risks between left-turning vehicles and straight-moving bicycles at signalized intersections, aiming to find suitable thresholds for vehicle-based early-warning systems. The authors utilized real-world traffic videos, extracting indicators such as PET, TTC, urgent braking actions, and trajectory uncertainty by reconstructing motion trajectories. A composite risk assessment framework was built to statistically explore relationships between these indicators and conflict situations. The findings suggest reasonable thresholds and combined alert conditions for risk identification, establishing a multi-stage warning mechanism and offering quantitative insights for designing effective vehicle warning systems in urban traffic.

(1) The paper used only camera data to reconstruct vehicle and bicycle trajectories, so accuracy might be limited, and it didn't incorporate other sensing technologies like LiDAR or radar, which could have improved precision.

(2) The study mainly relied on front-wheel positions when assessing risk, but in reality, vehicles and bicycles have different shapes and sizes, so ignoring these could oversimplify collision scenarios.

(3) The thresholds presented in the paper were calibrated only from a single intersection’s data, raising doubts about how broadly these thresholds can apply to other locations or situations.

(4) Human factors, like varying driver reaction times or cyclist behaviors, weren't sufficiently considered, so this oversimplification might not reflect real-world complexities.

(5) Trajectory uncertainty depends heavily on camera quality, video resolution, and manual tracking errors, which could have introduced additional inaccuracies into the risk assessment.

Author Response

We sincerely thank you for taking the time out of your busy schedule to carefully read and review our manuscript.
Your comments and questions are highly valuable and constructive, and they have greatly contributed to improving the quality of our work. We sincerely apologize for the insufficient explanations and descriptions in the original manuscript.
For your convenience, we have compiled our point‑by‑point responses to your valuable comments and suggestions into a PDF document, which has been attached to this submission.
We kindly invite you to review the attachment.

Best wishes
Shuncong Shen

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

  This manuscript presents a comprehensive study on the risk assessment of left-turn vehicle and bicycle conflicts at signalized intersections. The analysis of real-world traffic data through a composite risk index and the identification of risk thresholds for early warning systems is a significant contribution. However, the manuscript can be improved in terms of clarity, flow, and technical precision. The following detailed aspects provide specific suggestions for revision.

1. The introduction provides relevant background, but the flow can be improved. The connection between prior research and the study’s unique contributions should be made clearer. Consider rephrasing the sentence on the lack of analysis in real-world traffic scenarios and emphasize how your study fills that gap.
2. The transition to the proposed method in the introduction feels abrupt. Provide a brief paragraph linking the problem with the methods applied in this study.
3. The data collection section needs a clearer explanation of how the video data was synchronized and processed. It may be helpful to describe the data processing pipeline in a more structured way, perhaps with a flowchart. The terms "synchronized" and "preprocessed" should be more explicitly defined (e.g., the specific corrections made during preprocessing).
4. The descriptions of trajectory extraction and motion feature optimization are quite technical. Consider simplifying or breaking them down further for better readability.
5. The Kalman filter section is highly detailed but may overwhelm readers unfamiliar with the technical aspects. Summarize the key points more concisely and provide a reference for interested readers.
6. The author should highlight the rapid development of connected and autonomous vehicles in the background, thereby enhancing the research value of this work. Some recent work, such as “A MAS-Based Hierarchical Architecture for the Cooperation Control of Connected and Automated Vehicles, IEEE Transactions on Vehicular Technology, vol. 72, no. 2, pp. 1559-1573, Feb. 2023” can be cited.

Author Response

We sincerely thank you for taking the time out of your busy schedule to carefully read and review our manuscript.
Your comments and questions are highly valuable and constructive, and they have greatly contributed to improving the quality of our work. We sincerely apologize for the insufficient explanations and descriptions in the original manuscript.
For your convenience, we have compiled our point‑by‑point responses to your valuable comments and suggestions into a PDF document, which has been attached to this submission.
We kindly invite you to review the attachment.

Best wishes
Shuncong Shen

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

This research focuses on the analysis of conflicts between left-turning vehicles and straight-moving bicycles at signalized intersections in Japan. The objective is to improve in-vehicle early warning systems by developing a quantitative model to predict collision risks and define temporal and behavioral thresholds that can be used in real-time warning systems (HMI). Seven hours of video footage from a busy urban intersection in Tokyo were analyzed, identifying 37 conflict events with post-encroachment time (PET) ≤ 3 seconds. Six risk indicators were extracted, including PET (time between passages at the conflict point), TTC (time to collision), DRAC (deceleration required to avoid collision), Urgent Braking, Trajectory Variance (Kalman), and Passing Order (right-of-way). The study found that 81% of conflicts occurred with PET between 2–3 s, urgent braking occurred in 50% of cases with PET ≤ 2 s, and each 1 s reduction in PET increased the composite risk by 0.18 (R² = 0.55). The authors present a composite risk model (R), combining all indicators, and propose a two-level warning system: early warning (PET ≤ 2.5 s or TTC ≤ 1.5 s) and emergency warning (both criteria + urgent braking or high variance). The study provides a quantitative foundation for developing adaptive in-vehicle warning systems capable of anticipating conflicts with bicycles in complex urban environments, achieving up to 97% coverage of high-risk events with less than 6% false alarms.

The literature review is well-grounded, covering both traditional and recent approaches. It is up to date, with several sources from 2022 to 2025, and clearly justifies the study’s contribution by showing that there is still room for more integrated, data-driven risk models. However, it is recommended to clarify the research gap more objectively, i.e., the transition between previous studies and the article’s proposal should be more direct. It is also suggested to revise the text to reduce repetition, as the motivation for the study is repeated in several paragraphs. Additionally, it is recommended to include numbered specific objectives to facilitate reading.

The methodology is well described, with technical details and clear justifications. It is appropriate and innovative, combining computer vision, statistical modeling, and behavioral analysis, and has practical application potential in driver assistance and urban safety systems. However, it is suggested to explicitly indicate the software used in the statistical analysis (e.g., R, Python, SPSS), clearly separate sub-steps within each stage (e.g., section 2.3 could be divided into “trajectory extraction” and “coordinate transformation”), and include justification for the choice of thresholds (e.g., PET ≤ 2.5 s) based on literature or preliminary tests.

The presentation of results is consistent with the study’s objectives and methods, facilitates understanding of the findings even for non-specialist readers, and strongly supports practical recommendations for real-time warning systems. However, it is suggested to include p-values and confidence intervals in all relevant statistical analyses, improve figure readability with more descriptive captions (e.g., Fig. 7 could indicate the meaning of the symbols in the graph), and add a summary table with the main quantitative findings per indicator. Additionally, it is advisable to standardize the use of “Fig.” and “Figure,” as well as the use of bold for references to Figures and Tables. Figures 6 and 9 should be referenced in the body of the text before they appear.

The discussion includes a coherent interpretation of the results and well-defined practical applications (two-level HMI). However, it is suggested to include a more explicit comparison with previous studies (e.g., how do the results compare with those of Islam et al. or Guo et al.?), explore additional methodological limitations such as the influence of weather conditions or seasonal variations, and suggest improvements in experimental design for future studies (e.g., use of additional sensors, controlled simulations).

The conclusions summarize the main findings of this research, highlighting practical implications for road safety and warning systems, and pointing out limitations and future directions. However, the authors should avoid repeating phrases from the discussion and focus on synthesis and impact. It is recommended to explicitly mention how the results can be integrated into public policies or regulations.

The English in the article should be reviewed.

Author Response

We sincerely thank you for taking the time out of your busy schedule to carefully read and review our manuscript.
Your comments and questions are highly valuable and constructive, and they have greatly contributed to improving the quality of our work. We sincerely apologize for the insufficient explanations and descriptions in the original manuscript.
For your convenience, we have compiled our point‑by‑point responses to your valuable comments and suggestions into a PDF document, which has been attached to this submission.
We kindly invite you to review the attachment.

Best wishes
Shuncong Shen

Author Response File: Author Response.pdf

Reviewer 4 Report

Comments and Suggestions for Authors

The study provides a good level of detail regarding its methodology, which significantly aids in its potential replicability and reproducibility.

1. While software names are given, consider specifying the versions of the software used (e.g., Adobe Premiere Pro CC 2023, MATLAB R2023a, DIPP-Motion V X.X). This can sometimes affect specific functionalities or outputs and enhance reproducibility. 2. The "representative urban intersection" is described in terms of lane numbers and widths, but a more detailed schematic with precise real-world dimensions (e.g., lane widths, crosswalk length, turning radii, distances to traffic lights, camera mounting distances from specific intersection points) would significantly improve replicability. Figure 1 is useful but lacks these precise measured dimensions. 3. The study identifies 37 potential conflict events with PET ≤ 3 s. Elaborate slightly on the process of identifying these 37 events from the 7 hours of footage. Was it done manually by observers identifying near-misses, or was there an initial automated detection step based on proximity, followed by manual PET calculation and filtering? Clarifying this initial screening process would be beneficial. 3. If applicable, are statistical analyses, controls, sampling mechanism, and statistical reporting (e.g., P-values, CIs, effect sizes) appropriate and well described? The statistical analyses, sampling mechanism, and reporting are generally appropriate and well described. • Sampling: The study sampled 37 potential conflict events with PET ≤ 3.0 s from 7 hours of video data at a specific signalized intersection in Tokyo. The rationale for selecting the Itabashi district due to its high traffic accident rates is mentioned. • Analyses: The study utilizes regression analysis (OLS) to quantify the relationship between PET and the composite risk index (R). Nonparametric tests are mentioned, with a specific example being Welch's t-test for comparing risk levels between groups and Spearman correlation for assessing relationships. • Reporting: P-values are reported for the OLS regression (p < 0.001), Welch's t-tests (p < 0.0038), and Spearman correlation (p = 0.049). R² value (0.55) for the regression and standard deviations (SD) for group means are provided. Durbin-Watson and Shapiro-Wilk tests are reported for OLS assumptions.   Could you clarify the rationale for selecting the specific urgent braking thresholds (e.g., |𝑎_vehicle| > 3 m/s² or |𝑎_bicycle| > 2.5 m/s²)? Are these values derived from literature, a statistical analysis of acceleration distributions in your dataset, or another empirical basis? Explaining this choice would strengthen the justification for the "BrakeFlag" indicator.   While "nonparametric tests" are mentioned, explicitly name the specific non-parametric tests used where results are presented.   Consider a brief discussion in the limitations section about the generalizability of findings given the sample size (37 events) from a single intersection. While acknowledged, reinforcing this could guide future research more clearly.   Add a summary table detailing the composite risk index (R) components. Section 2.6 describes the six standardized proxy risk indicators (PET, TTC, vSum, DRAC, Dom, BrakeFlag) used in the R model. A table concisely defining each, including the specific formula or method of calculation (e.g., referencing Eqs. 3, 5, 6, 7), and stating their general relationship to risk (e.g., "smaller PET = higher risk") would greatly enhance the understanding of the R model. If weights (w_i) for the composite index were determined, including them would also be beneficial   Create a table summarizing the proposed warning thresholds and their performance. The discussion outlines a "union threshold" (PET ≤ 2.5 s or TTC ≤ 1.5 s) for early warning and "additional joint thresholds" (both PET ≤ 2.5 s and TTC ≤ 1.5 s, or urgent braking/DRAC surge, or Kalman speed variance > 1 m/s²) for emergency warning. A table consolidating these thresholds with their corresponding lead times, coverage rates (e.g., 95% of high-risk cases for primary warning, 97% for joint thresholds), and false alarm rates (under 6%) would be extremely useful for readers.   Enhance Figure 1 (Intersection layout) with numerical dimensions. As mentioned in the replicability section, adding specific measured dimensions (e.g., lane widths of 10.6m, pedestrian crosswalk width of ~4.6m) directly onto the diagram in Figure 1 would make it more informative and precise.   Clarify the "urgent braking doubled" statement: In the abstract, it is stated that "urgent braking doubled when PET ≤ 2 s"1. While the results show that 50% of cases with PET ≤ 2.0 s involved urgent braking, compared to 40% for PET > 2.5 s313, this is an increase but not a doubling. Please rephrase this statement for greater accuracy or provide the baseline percentage explicitly from which it "doubled." For instance, you could say "urgent braking increased by X% when PET ≤ 2 s" or "the rate of urgent braking increased from [baseline]% to 50% when PET ≤ 2 s."   Refine redundant purpose statement: The exact same sentence defining the purpose of the study appears in the abstract ("The purpose of this study is to quantify conflict situations between left-turning vehicles and straight-moving bicycles in real-world traffic environments, and to provide a foundation for determining the appropriate timing of future in-vehicle early warning systems.") and again at the very end of the Introduction section . Remove the repetition at the end of the Introduction. A strong concluding sentence summarizing the scope or approach of the current study, leading into the methods, would be more appropriate there.   In Table 4, the column headers for velocity and acceleration include arrows (e.g., 𝑣𝑣𝑒ℎ𝑖𝑐𝑙𝑒 (m/s) ↑). While these might intuitively suggest "increase" or "decrease" in the context of the data, they are not standard units or notations. Please clarify what these arrows signify in the table caption or a footnote to avoid ambiguity for readers.   While units are largely present, a general review to ensure all quantitative data across the entire paper consistently includes its unit at first mention or in tables/figures where appropriate would be beneficial. Comments on the Quality of English Language

Ensure consistent pluralization, for instance, "traffic accident" should be "traffic accidents"

There are occasional missing or superfluous articles. For example, "the specific interactions" could be "specific interactions". "converting files" instead of "converting the files" .

 Prepositions: "assess the risk for the distribution" could be "assess the risk from the distribution" or "assess the risk based on the distribution". "upon no deceleration" could be "with no deceleration" .

Wordiness/Phrasing: Some phrases could be more concise. "with the manual annotation of these contact points being conducted" could be "with manual annotation of these contact points conducted" . "because of dynamic scenarios" could be "due to dynamic scenarios"

Ambiguous Phrasing: The phrase "involved bicycles at the second part" is awkward and unclear. Based on the reference, it likely means "involved bicycles as the second party" in the accident. Clarifying this would be beneficial.

Line 92-93: "Shen et al. proposed the human-centered risk index Human-Centered Risk Index (HCRI)" has a redundant "human-centered risk index" preceding the full name.

Line 347-348: The statement "Welch’s t-test indicated a statistically significant difference in mean risk (𝑡 = 1.56, df ≈ 14, 𝑝 = 0.14)" may require further clarification. A p-value of 0.14 is not typically considered statistically significant at the conventional alpha level of 0.05. If the authors consider this significant, they should explicitly state the chosen significance level (alpha) for their analysis. Otherwise, the phrasing should be softened (e.g., "indicated a tendency towards a difference" or "showed a numerical difference").

Line 399-400: "two of the five events (60%) involving urgent braking" . 60% of five events is three, not two. Please double-check this value and correct either the count or the percentage.

Author Response

We sincerely thank you for taking the time out of your busy schedule to carefully read and review our manuscript.
Your comments and questions are highly valuable and constructive, and they have greatly contributed to improving the quality of our work. We sincerely apologize for the insufficient explanations and descriptions in the original manuscript.
For your convenience, we have compiled our point‑by‑point responses to your valuable comments and suggestions into a PDF document, which has been attached to this submission.
We kindly invite you to review the attachment.

Best wishes
Shuncong Shen

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

  The author has addressed my concerns well. I would recommend the publication of this paper. However, the author should add more descriptions to highlight the core innovation of this work. Furthermore, the rapid development of autonomous vehicles can refer to the recent work “J. Liang, K. Yang, C. Tan, J. Wang and G. Yin, Enhancing High-Speed Cruising Performance of Autonomous Vehicles Through Integrated Deep Reinforcement Learning Framework, IEEE Transactions on Intelligent Transportation Systems, vol. 26, no. 1, pp. 835-848, Jan. 2025”.

Author Response

Dear Dr. Reviewer

We sincerely thank you for taking the time out of your busy schedule to carefully read and review our revised manuscript. We truly appreciate your recognition and encouragement for our work, as well as the valuable comments and suggestions you provided.
Your feedback has been extremely helpful in improving the quality of our paper. We will continue to refine our research in future work and strive to make further contributions to the field.

For your convenience, we have compiled our point‑by‑point responses to your valuable comments and suggestions into a PDF document, which has been attached to this submission, and we kindly invite you to review the attachment.

Once again, thank you very much for your thoughtful review and support.
Best regards,

Author Response File: Author Response.pdf

Reviewer 4 Report

Comments and Suggestions for Authors

same to editor

Author Response

Dear Dr. Reviewer

We sincerely thank you for taking the time out of your busy schedule to carefully review our revised manuscript.
We truly appreciate your recognition and encouragement for our work, as well as the valuable comments and suggestions you provided. Your feedback has been extremely helpful in improving the quality of our paper.

We will continue to refine our research in future work and strive to make further contributions to the field.

Once again, thank you very much for your thoughtful review and support.
Best regards,