Intersection Safety Assessment Using Video-Based Traffic Conflict Analysis: The Case Study of Thailand

Round 1

Reviewer 1 Report

This is a very interesting analysis of safety assessment of intersections using video-based traffic conflict analysis. The authors have provided some very interesting results; however, there are some clarifications that should be made to help improve the manuscript. These are provided as follows:

The authors have provided a wealth of information from the video-based analysis but do not provide any application. How can the reader take the results that have been presented here and apply them in their own situation? You note that the results are somewhat intuitive, so if that is the case, what is the contribution? How can we use these?

Where are your study sites located (country)? There are A LOT of motorcycle crashes in the data. Are the results applicable in areas with far fewer motorcycles? How does this play into the analysis and results?

You talk on page 8 about deep learning algorithms used to train and analyze the data. How do these work? There is not much detail on this.

It is noted in the paper that this analysis can be used in place of using limited crash data. Have you taken that limited crash data though and tried to correlate it to the video conflict data to identify correlations or trends? If not, how do you know that it possibly better reflects the real crash risk (line 413)?

Grammatical/Editorial comments:

Line 18 - Change "applicable" to "apply"

Line 19 - The word "the" before "risk-based" is not appropriate.

Line 52 - Remove "the" before "safety analysis." Grammatical changes like this need to be addressed throughout the document.

Acronyms - The authors have defined and redefined (over and over) the acronyms in the document. Define acronyms on first use and then use those acronyms consistently throughout the remainder of the document.

Line 137/138 - Refer to equations by number throughout.

Line 149 - Change "detect" to "detecting"

Line 159 - This sentence does not make sense and runs into the next sentence. Look at rewording.

Line 162 and 167 - The reference to the authors in these two references is confusing. Either the authors are not listed correctly in the References (by the last name) or the first names are used here. Please double-check all references and author names to ensure that they are referenced accordingly.

Line 199 - Change to "A total of 12 intersection sites were examined."

Line 217 - The phrase "which be required more than one camera..." does not make sense.

Line 270 - Again, on the author name, is this the last name (Tran)? If so, the reference is not listed by the last name. It makes me wonder how many others are incorrect as well. There are only a select few where you reference the author by name in the text and most of those seem to be inconsistent or incorrect.

Line 280 - What is meant by "probabilities of fatality"?

Line 311 - You are inconsistent in how you note the evening hour (p.m. vs. P.M.) in the document. Be consistent.

Line 338 - When you say "highly related" do you mean "correlated"?

I did not review the References in detail but I am concerned about the inconsistencies that were noted previously. These need to be addressed.

Author Response

We appreciate the valuable comments from the Reviewer. We have revised our manuscript accordingly, and have summarized our replies to the Reviewer's comments. 

Author Response File: Author Response.pdf

Reviewer 2 Report

An interesting paper on an important topic, which shows how surrogate safety measures can be successfully applied in road safety research.

My major concern is that the speed of conflicting vehicles is considered twice in this methodology, that is while calculating the Time to Collision and, separately, while considering the vehicle speeds for the risk calculation. In my opinion, there could be a disproportionate emphasis on high speeds for high-risk conflicts, such as it appears for arterial-arterial conflicts. However, usually, especially if traffic controlled, these intersections may be even safer (in terms of crash rates, considering traffic volumes) than low-level intersections. Hence, I suggest to thoroughly explain and discuss these issues (especially by justifying the use of speeds for calculating the risk index), also possibly comparing results with crash-based safety research literature.

Other comments are as follows:

• why the collector/collector intersection is not considered?
• there is a high percentage of motorcycles (especially on local roads) which can be hardly found in some other countries. Please discuss transferability issues.
• Bicycles are not considered in the traffic analysis. Please explain.
• Trucks are recorded in the traffic analysis but they are not considered in the conflict analysis. Please explain.
• The crossing and rear-end conflicts are considered (the "merging" conflict is not taken into account). However, this difference is not used later in the paper (while discussing results and elaborating on different risks). It could be interesting to take into account these differences.
• Equation 2 is used to calculate percentages, however sentences above the equation are not clear (is it used for high risk collisions or for all risky collisions?)
• I suggest to improve the discussion section by focusing on differences between intersection and conflict types.

Author Response

We appreciate the valuable comments from the Reviewer. We have revised our manuscript accordingly, and have summarized our replies to the Reviewer's comments. 

Author Response File: Author Response.pdf