Round 1

Reviewer 1 Report

This paper used a Bayesian network model to mine the complex relationship between 17 unsafe behaviors and sideswipe collisions, quantify the influence of each unsafe behavior and identify the chain of unsafe behaviors. This paper is relatively complete but suffers from the following problems.

1        The abstract section lacked a specific presentation of the study results. So does the title.

2        The review section of the paper did not have a review related to the training methods for small sample training sets. Therefore, the conclusion in line 79 of the paper lacked basis.

3        The video selection criteria in lines 98 to 99 lacked specific descriptions, as the purposeful selection is known to affect the accuracy of the analysis results.

4        C13 Distracted and inattentive driving. How this feature was observed.

5        The method section of the paper lacked specific computational principles for the derivation of the model and needed additional explanation. Also, the structure of that section needs to be adjusted, and I think you can refer to the way these papers are written.

[1] Liu L, Ye X, Wang T, Yan X, Chen J, Ran B. Key Factors Analysis of Severity of Automobile to Two-Wheeler Traffic Accidents Based on Bayesian Network. International Journal of Environmental Research and Public Health. 2022; 19(10):6013. https://doi.org/10.3390/ijerph19106013

[2] Xiaofei Ye, Yi Zhu, Tao Wang, Xingchen Yan, Jun Chen, Bin Ran. Level of Service Model of the Non-Motorized Vehicle Crossing the Signalized Intersection Based on Riders’ Perception Data. International Journal of Environmental Research and Public Health. 2022; 19 (8):4534.

[3] Chen, H.; Zhao, Y.; Ma, X. Critical factors analysis of severe traffic accidents based on Bayesian network in China. J. Adv. Transp. 2020, 2020, 8878265.

6        What is the basis for the 6 factors identified in line 252 as influencing the occurrence of sideswipe collisions, and if only the threshold value, please explain what exactly is meant by the threshold value, also it seems that the paper lacked a specific formula for calculating the Predictive variable importance in Figure 3.

7        Lines 259-261 mentioned that the unsafe behaviors that affect the occurrence of sideswipe collisions were all related to irregular behavior and inattention when changing lanes, but feature C13 (Distracted and inattentive driving) is not in the influencing factors, I think this is something that needs to be discussed here.

8        As far as I understand, the output of a Bayesian network model after inputting a set of sample values is a set of probabilities, and in the context of this paper the output should be the probability of a collision occurring and the probability of no collision occurring. So in this output format, how is the accuracy of the model calculated in table2 and section 4.1?

9        What is the role of the six factors identified in the Figure 3 section of this paper, and why not eliminate the features with less influential factors to train subsequent models?

10      What are the criteria for removing part of the chain of unreasonable behavior in line 329 of the paper?

11      As shown in Figure 7, when no collision occurs, why did the possibility of C11 (Failure to turn on signal when changing lanes) happening increase instead (62.3% to 94.4%)? It is suggested that the authors add relevant analyses.

12      The analysis in section 4.4 of the paper is good and of practical value.

Author Response

Response to Reviewer 1 Comments

• The abstract section lacked a specific presentation of the study results. So does the title.

I agree with the reviewer's opinion. The modified part has been marked in the paper. The description section has been modified. The abstract section added a specific presentation of the study results, the additions are as follows. The significant influential single unsafe behavior leading to sideswipe collision on urban expressways was lane change without checking the rearview mirror or not scanning the road around and queue-jumping. Moreover, based on unsafe behavior chains analysis, the most influential chains leading to sideswipe collision were: improper driving behavior in an emergency−failure to turn on signal when changing lanes−distracted and inattentive driving. The newly revised title of article is “Unsafe behaviors analysis of sideswipe collision on urban expressways based on Bayesian network”.

• The review section of the paper did not have a review related to the training methods for small sample training sets. Therefore, the conclusion in line 79 of the paper lacked basis.

I agree with the reviewer's opinion. The description of the paragraph has been revised and is marked in the text.

3        The video selection criteria in lines 98 to 99 lacked specific descriptions, as the purposeful selection is known to affect the accuracy of the analysis results.

The video selection criteria include time limit, place of occurrence limit, video process, video quality these four requirements. According to the crash video, the selected crashes occurred in the main road and ramp area of several urban expressways of Hefei during the period from 2016 to 2020. Selected crash videos could reflect the whole process of the crash, including drivers’ unsafe behaviors before the crash and the occurrence of the crash. The images in the video were clear and effective. Besides, they could identify unsafe behaviors during the whole process of a crash. For example, for a sideswipe collision, the research process is the period from the moment when a vehicle shows the lane change trend to the moment when a crash occurs, which is an independent research sample. For non-crash videos, the research sample starts when a vehicle has a tendency to change lanes, and it ends when a sideswipe collision is likely to occur, that is, the moment when the transverse distance between vehicles is minimum.

Video selection has no purpose. All videos that meet the above video requirements can be used as research samples. All video data are the natural driving data of the driver in the normal driving process, excluding purposeful video selection.

Crash videos were obtained from the public security traffic police department of Hefei in China. The crash videos came from the driving data recorder installed in the vehicle by the driver himself, which was predominantly used to record the crash process. Once the crash result is disputed, it is convenient for the traffic police to determine the responsibility of the collision. The video is to identify those responsible for the crash, not research. In this study, I used relatively new data set for the study of unsafe behaviors in crashes, which is also one of the innovations of this paper.  

4        C13 Distracted and inattentive driving. How this feature was observed.

C13 (distracted and inattentive driving) is selected for this unsafe behavior should be contributed to the occurrence of urban expressway crashes.  First, the types of unsafe behaviors listed are important in causing crashes. Besides, unsafe behaviors that can be identified from the video. The video is taken from the driver's dashcam. That is, from the driver's first perspective, the unsafe behavior that can be identified and classified as the research object of this topic. Based on this, C13 is defined.

C13, distracted and inattentive driving (Zhang et al. 2020). Distracted and inattentive driving refers to behavior leading to missed observations of some kind, which in turn leads to a critical event of ‘timing’ (premature, late action, or no action) or (incorrect) ‘direction’. When the driver lacks motivation to carry out their task in the best way possible, an object or sequence of events diverts the driver’s attention, or the driver is used to ascertaining the environment makes it difficult to discover changes (Talbot et al, 2013). The specific performance of this behavior includes the driver’s failure to reduce speed in time or improper driving behavior in an emergency. The discriminant method of driver’s failure to reduce speed in time is that, accounting for possible unfavorable driving situation, the vehicle maximum deceleration is set for 10 m/s² and 8 m/s² for dry and wet roads, respectively (Xiong et al. 2019). When the vehicle starts to decelerate until it stops, the maximum deceleration does not reach the value (Xiong et al. 2019). The vehicle is considered to fail to reduce speed in time. Improper driving behavior in an emergency is another form of distracted and inattentive driving. This unsafe behavior means that a driver improperly operates the vehicle when a vehicle in front changes a lane, which leads to more vehicles being involved in a crash.

5       The method section of the paper lacked specific computational principles for the derivation of the model and needed additional explanation. Also, the structure of that section needs to be adjusted, and I think you can refer to the way these papers are written.

[1] Liu L, Ye X, Wang T, Yan X, Chen J, Ran B. Key Factors Analysis of Severity of Automobile to Two-Wheeler Traffic Accidents Based on Bayesian Network. International Journal of Environmental Research and Public Health. 2022; 19(10):6013. https://doi.org/10.3390/ijerph19106013

[2] Xiaofei Ye, Yi Zhu, Tao Wang, Xingchen Yan, Jun Chen, Bin Ran. Level of Service Model of the Non-Motorized Vehicle Crossing the Signalized Intersection Based on Riders’ Perception Data. International Journal of Environmental Research and Public Health. 2022; 19 (8):4534.

[3] Chen, H.; Zhao, Y.; Ma, X. Critical factors analysis of severe traffic accidents based on Bayesian network in China. J. Adv. Transp. 2020, 2020, 8878265.

It has been modified and improved according to expert opinions.

6       What is the basis for the 6 factors identified in line 252 as influencing the occurrence of sideswipe collisions, and if only the threshold value, please explain what exactly is meant by the threshold value, also it seems that the paper lacked a specific formula for calculating the Predictive variable importance in Figure 3.

Fig. 1. The distribution diagram of unsafe behaviors on crash videos.

According to the original statistical data results as shown in Fig. 1. It can be seen that once sideswipe collisions occur on the urban expressway, most of the unsafe behavior number in each collision are concentrated in 2-5, and one unsafe behavior proportion is 14.3%.

Although most sideswipe collisions occur in the form of collision chain, the single unsafe behavior has a certain influence on sideswipe collisions, and the influence can not be ignored.  Therefore, I listed the impact value of a single unsafe behavior on the sideswipe collision in the form of Fig. 2.

The original article lists six single unsafe behaviors with relatively high impact and does not set thresholds. The following 11 unsafe behaviors have an impact on the sideswipe collision, but have less impact. While driving on the expressway, drivers should also pay enough attention to it. The original text has been modified to describe the effects of the 17 unsafe behaviors.

The modified contents will be: Although most sideswipe collisions occur in the form of collision chain, the single unsafe behavior has a certain influence on sideswipe collisions, and the influence can not be ignored. Of all the single factors，C17 (lane change without checking the rearview mirror or not scanning the road around), C12 (queue-jumping), C8 (unsafe passing), C16 (improper driving behavior in an emergency), C11 (failure to turn on signal when changing lanes), and C9 (unsafe merging) affect the occurrence of sideswipe collision relatively obviously.​ And other unsafe behaviors have a certain influence on the occurrence of sideswipe collision on urban expressways, and their impacts are smaller than those of the top six unsafe behaviors. Reducing every unsafe behavior will play a positive role in reducing collisions. Thus, regulating drivers’ driving behaviors is particularly important in avoiding crashes.

Fig. 2. The variable importance bar chart.

7        Lines 259-261 mentioned that the unsafe behaviors that affect the occurrence of sideswipe collisions were all related to irregular behavior and inattention when changing lanes, but feature C13 (Distracted and inattentive driving) is not in the influencing factors, I think this is something that needs to be discussed here.

I agree with the experts' opinions. C13 does appear frequently on urban expressways and is one of the most common unsafe behaviors.

When I analyzed the impact of unsafe behaviors on collisions, the statistical analysis showed that C13, such unsafe behaviors, often appeared in rear-end collisions. When the vehicle driving in the same lane of the expressway is distracted, it will collide with the vehicle in front, and then cause a rear-end collision on the expressway.

In sideswipe collisions, drivers will have purposeful lane change behavior. This kind of collision caused by distraction is not common, but it also happens occasionally. Therefore, it is not the most common and main unsafe behavior that causes sideswipe collisions on expressway. 

In order to better explain the C13 that causes sideswipe collisions, I have defined C13 in this paper.

C13, distracted and inattentive driving. Distracted and inattentive driving refers to behavior leading to missed observations of some kind, which in turn leads to a critical event of ‘timing’ (premature, late action, or no action) or (incorrect) ‘direction.’ When the driver lacks motivation to carry out their task in the best way possible, an object or sequence of events diverts the driver’s attention, or the driver is used to ascertaining the environment makes it difficult to discover changes.

8        As far as I understand, the output of a Bayesian network model after inputting a set of sample values is a set of probabilities, and in the context of this paper the output should be the probability of a collision occurring and the probability of no collision occurring. So in this output format, how is the accuracy of the model calculated in table2 and section 4.1?

Classification index evaluation

In the classification task, the calculation basis of each indicator comes from the classification results of positive and negative samples, which can be expressed by the confusion matrix: 

Table 1. Confusion matrix.

	Predicted crashes	Predicted non-crashes
Real crashes	Tcrash	Fnon_crash
Real non-crashes	Fcrash	Tnon_crash

Table 2. Indicator description

Item	Description	Correct or not
Tcrash	The actual number of positive samples is predicted to be positive samples	Correct
Tnon_crash	The actual number of negative samples is predicted to be negative samples	Correct
Fnon_crash	The actual number of positive samples is predicted to be negative samples	False
Fcrash	The actual number of negative samples is predicted to be positive samples	False
Tcrash+ Fcrash	The predicted number of positive samples
Fnon_crash+ Tnon_crash	The predicted number of negative samples
Tcrash+ Fnon_crash	The actual number of positive samples
Fcrash+Tnon_crash	The actual number of negative samples

The main indicator of classification index evaluation includes as shown below:

1. Overall accuracy

 Definition：(Tcrash+Tnon_crash)/(Tcrash+Fnon_crash+Fcrash+Tnon_crash)

2. The false alarm rate

 Definition：(Tnon_crash)/(Fcrash +Tnon_crash)

3. Precision

 Definition：(Tcrash)/(Tcrash+Fcrash)

4. Recall

 Definition：(Tcrash)/(Tcrash+Fnon_crash)

Regression index evaluation

The evaluation index of classification problem is accuracy, so the evaluation index of regression algorithm is MSE, RMSE, MAE and R-Squared. In the formula,  is the actual value, and the is the predicted value, n is the number of samples. 

The main indicator of regression index evaluation includes as shown below:

1. MSE

MSE, Mean Squared Error.

Definition:

2.RMSE

RMSE, root mean squard error.

Definition:

3.MAE

MAE, mean absolute deviation.

Definition:

4.MAPE

MAPE, Mean Absolute Percentage Error.

Definition:

In this paper, through the prediction of unsafe behavior for collisions, the result is a classification index, that is, to judge whether there is a sideswipe collision.

Based on the videos data collected on the urban expressway to test, two models were developed using Bayesian networks. 17 unsafe behaviors were the influencing factors, and whether sideswipe collisions occurred was the output result. The results obtained by Bayesian network prediction were shown in the figure below. According to the comparison between the prediction results of the BN model and the statistical calculation results of the original testing datasets, the learning error of the BN model can be obtained to evaluate the accuracy of the model. This paper used the classification index evaluation index to evaluate the prediction results of the model. In addition to the accuracy, precision, recall three indicators, other indicators were listed in the paper. We could get the accuracy, precision, recall, the false alarm rate, these four indicators

Fig. 3. A partial presentation of the predicted results

9        What is the role of the six factors identified in the Figure 3 section of this paper, and why not eliminate the features with less influential factors to train subsequent models?

Figure 3 of the paper is shown below. The figure shows the influence of single unsafe behavior on sideswipe collisions of expressway is demonstrated.

It can be seen that once sideswipe collisions occur on the urban expressway, most of the unsafe behaviors in each collision are concentrated in 2-5, and only one unsafe behavior proportion is 14.3%.

And other unsafe behaviors have a certain influence on the occurrence of sideswipe collision on urban expressways, and their impacts are smaller than those of the top six unsafe behaviors.

Fig. 4.  The variable importance bar chart.

10      What are the criteria for removing part of the chain of unreasonable behavior in line 329 of the paper?

​        According to the original statistical data results as shown in Fig. 5. It can be seen that once sideswipe collisions occur on the urban expressway, most of the unsafe behaviors in each collision are concentrated in 2-5, and only one unsafe behavior proportion is 14.3%. According to data statistics, the number of unsafe behaviors in a sideswipe collision is not more than 5. Therefore, the chain of more than five unsafe behaviors obtained by Bayesian network analysis was removed in this paper, which was inconsistent with the actual traffic collision situation. I removed the behavior chain that existed in theory but did not appear in the actual situation.

Fig. 5. The distribution diagram of unsafe behaviors on crash videos.

11      As shown in Figure 7, when no collision occurs, why did the possibility of C11 (Failure to turn on signal when changing lanes) happening increase instead (62.3% to 94.4%)? It is suggested that the authors add relevant analyses.

Through the investigation of sideswipe collisions on urban expressway, as well as the data investigation of reference unsafe behavior, but no real collision. ​C11 (failure to turn on signal when changing lanes) is a very common behavior, in the analysis of statistical data, this unsafe behavior accounted for a large proportion.

C11 (failure to turn on signal when changing lanes) is needed in the following case, a driver should turn on the turn signal in advance when he makes turns, changes lanes, prepares to overtake, or leaves a parking place. 

The occurrence of this behavior will not necessarily lead to the occurrence of sideswipe collisions. The occurrence of collisions is not only related to the driver's own unsafe behaviors, but also affected by the surrounding vehicles and the surrounding traffic environment.

The driver needs to turn on his signal when he enters the main road. However, the presence of C11 (failure to turn on signal when changing lanes) does not necessarily cause a collision when there are no or few vehicles around the lane change vehicle that needs to enter the main road. This also applies to the situation that the vehicle makes turns, changes lanes, prepares to overtake, or leaves a parking place. 

However, when there is C11 (failure to turn on signal when changing lanes) and other unsafe behaviors, that is, a series of unsafe behaviors, there is still a certain probability of sideswipe collisions. 

As shown in Table 5 in the origin paper, C16 (improper driving behavior in an emergency) =yes; C11 (failure to turn on signal when changing lanes) =yes; ​the probability is 0.8. It means that the driver does not turn on the turn signal when changing lanes, and a vehicle is affected by the lane change does not brakes in time, but chooses to turn to other lanes, that is, C16 (improper driving behavior in an emergency) occurs, and other vehicles will be involved in a collision, that is, a multi-vehicle sideswipe collision happens. 

Therefore, in this paper, when no collision occurs, the possibility of C11 (failure to turn on signal when changing lanes) happening increases. It means that although the frequency of C11 (failure to turn on signal when changing lanes) is high, it does not necessarily cause collisions when it appears alone. However, when combined with other unsafe behaviors, it is one of the factors promoting the occurrence of collisions. Therefore, when the collision is inevitable, it is one of the influencing factors, accounting for the risk proportion of 0.623. When no collision occurs, the possibility of C11 (failure to turn on signal when changing lanes) is 0.944, meaning that it is very common, accounting for a high proportion, but the impact on the collision is not very significant.

Table 3. Unsafe behavior description

Number	unsafe behavior	Number of unsafe behaviors in sideswipe collision	Number of unsafe behaviors in non-collision
C1	Exceeding Authorized Speed Limit	3	0
C2	Improper Parking	0	5
C3	straddling lanes without changing lane	3	12
C4	Competitive/competitive/aggressive driving	7	1
C5	Lane departure/straddling lanes	11	59
C6	Got involved in unofficial races	5	25
C7	Close following (<1 s)	1	1
C8	Unsafe passing	9	4
C9	Unsafe merging	14	4
C10	Forget to put the hazards lights on	0	6
C11	Forget to signal when changing lane	48	509
C12	Queuing, nearly hit the car in front	21	19
C13	Distracted & inattentive driving	1	2
C14	Tailgating to force others to give way	6	3
C15	Stop longer than expected when braking	0	1
C16	Wrong gear in case of emergency	10	0
C17	Manoeuvre without checking mirror or not scanning road around	40	12

 

12      The analysis in section 4.4 of the paper is good and of practical value.

Thank the experts for their affirmation.

References

[1] Talbot R., Fagerlind H., Morris, A., (2013). Exploring inattention and distraction in the SafetyNet Crash Causation Database. Accident Analysis and Prevention, 60, 445–455.

[2] Xiong X.X., Wang M., Cai Y.F., Chen L., Farah H., Hagenzieker M., (2019). A forward collision avoidance algorithm based on driver braking behavior. Accident Analysis and Prevention, 129, 30–43.

[3] Yang L.P., Feng Z.X., Zhao X.H., Jiang K., Huang Z.P., (2020). Analysis of the factors affecting drivers’ queue-jumping behaviors in China. Transportation Research Part F, 72, 96–109.

[4] Zhang Z., Guo Y.S., Fu R., Yuan W., Wang C., (2020). Linking executive functions to distracted driving, does it differ between young and mature drivers? Plos One, 15(9), e0239596.

 

 

 

 

 

 

 

 

 

 

 

Author Response File: Author Response.pdf

Reviewer 2 Report

Dear authors,

I enjoyed reading your manuscript. I appreciate the connection of the Bayesian network with drives' unsafe behavior focused on sideswipe collision.

I have only one minor comment. You had to analyze a large number of records. However, I would like to ask one question: You wrote that there were 605 video samples analyzed in the whole article. But in chapter 2.2. you mentioned that there were 663 unsafe behavior samples.

I also want to ask what was the total length (time) of all analyzed video files. How was it possible to investigate this amount of material? How many people performed this raw analysis? This time-consuming process should be shortly described in the article.

Thank you.

Author Response

Response to Reviewer 2 Comments

1.I have only one minor comment. You had to analyze a large number of records. However, I would like to ask one question: You wrote that there were 605 video samples analyzed in the whole article. But in chapter 2.2. you mentioned that there were 663 unsafe behavior samples.
        The number of videos and unsafe behaviors has been described in the paper. The analysis data included 605 videos, including 70 sideswipe collision videos and 535 non-crash videos. There were 179 unsafe behaviors in the 70 crash videos. And, there were 663 unsafe behaviors in the 535 non-crash videos.
2.I also want to ask what was the total length (time) of all analyzed video files. How was it possible to investigate this amount of material? How many people performed this raw analysis? This time-consuming process should be shortly described in the article.
        The total time of 70 sideswipe collision videos is 452 seconds. The total time of 535 non-crash videos is 7500 seconds.

Members of the research team spent a year sorting videos related to unsafe behaviors. The process of collating data involves:

(1) A collection of original arrangements.

(2) Analyze collision types.

(3) Determine the types of unsafe behaviors and the classification criteria for each unsafe behavior.​ (4) Calculate the duration of each crash, including the duration of non-crashes. According to the vehicle identification in the video, calculates the time interval between two adjacent frames. The video was played frame by frame, the number of frames per second of images was calculated and determined many times, and the interval time between two adjacent images was calculated at the same time.

(5) Analyze the type and number of unsafe behaviors displayed by video during each crash and non-crash.

(6) Check each crash and non-crash. One of the advantages of the crash video is that we can watch it repeatedly to check the correct parameter values. The data sets in this paper are checked repeatedly to ensure the correctness and rationality of the analysis.

Selected crash videos could reflect the whole process of the crash, including drivers’ unsafe behaviors before the crash and the occurrence of the crash. Images in the video were clear and effective, and above all, they could identify unsafe behaviors during the whole process of a crash. For example, for a sideswipe collision, the research process is the period from the moment when a vehicle shows the lane change trend to the moment when a crash occurs, which is an independent research sample. For non-crash videos, the research sample starts when a vehicle has a tendency to change lanes, and it ends when a sideswipe collision is likely to occur, that is, the moment when the transverse distance between vehicles is minimum.

 

 

 

 

 

 

 

 

 

 

Reviewer 3 Report

The title should be changed. If we investigate and use the word impact, it should be „impact of something on something”. Now we have the only impact of something.

Why in the introduction does the author show old data from 2017? Please use newer.

There is no related literature chapter with other methods. Authors should show what was done in the literature and what is the new part in their work.

How do authors, based on the video, calculate the speed of vehicles?

How do authors, based on the video, calculate the acceleration of vehicles?

queue-jumping has different definition: A queue jump is a type of roadway geometry used to prefer buses at intersections, often found in bus rapid transit systems.

Authors should contain information about weather conditions during that crash. It could have a huge impact.

Please edit fig 1,2 to make it more readable

Fig 1 and 2 has five unsafe behaviors, so why do the authors define 17?

What is figure 1 and 2 is unsafe number 1? Before authors define is as C.

 

Please add scheme for method.

Author Response

Response to Reviewer 3 Comments

1.The title should be changed. If we investigate and use the word impact, it should be „impact of something on something”. Now we have the only impact of something.

The newly revised title of article is “Unsafe behaviors analysis of sideswipe collision on urban expressways based on Bayesian network”.

2.Why in the introduction does the author show old data from 2017? Please use newer.

The statistics of vehicle collisions in China in 2020 have been updated according to experts' opinions. According to the latest data of the Statistical Yearbook of the People's Republic of China, it is up to 2021, and the latest data only update the national road traffic accident data in 2020, so the latest national accident data in 2020 is used.

3.There is no related literature chapter with other methods. Authors should show what was done in the literature and what is the new part in their work.

This part has been modified in the paper. At present, research on traffic safety problems using the Bayesian method is mostly carried out from the perspective of various influencing factors of collisions. There is no special research on drivers' behavior factors using the Bayesian method using crash videos. Besides, this paper focused on the influence of driver's behavior on the sideswipe collisions on urban expressway, not only focusing on unsafe behaviors, but also the chain combination among unsafe behaviors.

4.How do authors, based on the video, calculate the speed of vehicles?

First, calculate the speed according to the reference in the video or get the speed directly from the dashboard recorded. In order to get the speed of vehicle from the reference in the video, the specific method is as follows.

(1) Choose a reference from the video. According to the identification of vehicle driving conditions in the video, relatively obvious and easy to measure references are selected, such as roadside lampposts, the white dividing line, and central dividing lines on the road.  

(2) Calculates the time interval between two adjacent frames. The video was played frame by frame, the number of frames per second of images was calculated and determined many times, and the interval time between two adjacent images was calculated at the same time.  

(3) Calculate the distance between 2 references.  Use image processing software tools to measure the distance between two references.  

(4) Select the point of spatial coincidence. Play video frame by frame, select a feature point on the vehicle (such as the front of a vehicle), mark the feature point as frame 1 when it passes the first reference, and mark it as frame n when the front passes the second reference, and then calculate the number of frames between the two images.  

(5) Speed calculation.  The speed is calculated according to the formula of velocity, time, and displacement in mechanics.  

5.How do authors, based on the video, calculate the acceleration of vehicles?

First, calculate the speed according to the reference in the video or get the speed directly from the dashboard recorded. And then a vehicle acceleration rate or a vehicle deceleration rate may be obtained based on the speed calculated before. The specific method is as follows.

(1) Judge vehicle motion state. In the actual speed calculation work, according to the video image of the moving vehicle to make a preliminary judgment. We can determine the state of the vehicle, whether the vehicle is accelerating or decelerating can be analyzed.

(2) Choose a reference from the video. According to the identification of vehicle driving conditions in the video, relatively obvious and easy to measure references are selected, such as roadside lampposts, the white dividing line, and central dividing lines on the road.  

(3) Calculates the time interval between two adjacent frames. The video was played frame by frame, the number of frames per second of images was calculated and determined many times, and the interval time between two adjacent images was calculated at the same time.  

(4) Calculate the distance between 2 references.  Use image processing software tools to measure the distance between two references.  

(5) Select the point of spatial coincidence. Play video frame by frame, select a feature point on the vehicle (such as the front of a vehicle), mark the feature point as frame 1 when it passes the first reference, and mark it as frame n when the front passes the second reference, and then calculate the number of frames between the two images.  

(6) Speed calculation.  The speed is calculated according to the formula of velocity, time, and displacement in mechanics.  

(7) The acceleration between two adjacent points in the calculation. In this method, it is assumed that the vehicle is in uniformly variable motion between two adjacent frames. And the acceleration is introduced to establish a time interpolation calculation model considering the acceleration between frames.  

                         (1)

Where, ai is the acceleration of the vehicle. Vi is the velocity of feature point 1. The time corresponding to this velocity is ti. Vi+1 is the velocity of feature point 2. The time corresponding to this velocity is ti+1.

6.queue-jumping has different definition: A queue jump is a type of roadway geometry used to prefer buses at intersections, often found in bus rapid transit systems.

Yang et al. (2020) referred to queue-jumping behavior is that the driver of the subject vehicle forcibly merges into the target lane on the condition that the lane-changing gap is insufficient. This behavior usually takes place in the traffic situation of vehicle queueing. This behavior is often particularly severe at some traffic bottlenecks (such as intersections and merging and diverging areas of expressways and especial congestion sections), as shown in figure 1 below. Besides, the paper in order to better explain this kind of unsafe behavior, the corresponding rules of queue-jumping behavior are given. The initial horizontal distance between a lane-changing vehicle and a straight-moving vehicle is less than 2.2 m. The maximum lateral acceleration of the straight-moving vehicle is less than 0.07 g, and lane offset is less than 1.0 m. The maximum length of lane-change process is no more than 75 m. The velocity of both lane-changing and straight-moving vehicles should be more than 1 m/s.

Fig. 1. A schematic diagram of queue-jumping at an expressway exit.

7.Authors should contain information about weather conditions during that crash. It could have a huge impact.

I also agree with the expert about the impact of weather on collisions. Weather does have an impact on collisions, which has been discussed in many traffic safety literatures. In this paper, due to the limitation of collision location, it is difficult to collect only collision videos on urban expressways, and the total number of accident videos is 70 sideswipe collision videos.  If the discussion is classified according to weather factors, the results may be biased due to insufficient sample size. At the same time, this paper focuses on the analysis of the impact of unsafe behavior on the collisions. Therefore, this paper does not consider other factors too much.  The influence of unsafe behaviors on collisions under different weather conditions can be further discussed as a topic of research.  

8.Please edit fig 1,2 to make it more readable

The modified part has been marked in the paper. The description section has been modified.

Fig. 1.  The distribution diagram of unsafe behaviors on crash videos.

Fig. 2. The distribution diagram of unsafe behaviors on non-crash videos.

9.Fig 1 and 2 has five unsafe behaviors, so why do the authors define 17?

Figure 1 describes the distribution diagram of unsafe behaviors on crash videos. The horizontal axis shows the number of unsafe behaviors. It is clearly marked in the drawing. So the number is 5, instead of 17.

Figure 2 shows the distribution diagram of unsafe behaviors on non-crash videos. The horizontal axis shows the number of unsafe behaviors. It is clearly marked in the drawing. So the number is 5, instead of 17.

10.What is figure 1 and 2 is unsafe number 1? Before authors define is as C.

Figure 1 and 2 have been modified to represent the distribution of unsafe behavior in collision. In the following discussion, these 17 types of unsafe behaviors are classified. In order to facilitate classification, they are represented by C. C stands for category.​

11.Please add scheme for method.

This part has been added to the article.

Take the TAN as an example, explain the steps of structure learning algorithm are as follows:

(1) Calculate the conditional mutual information of all input variables Xi and Xj. The conditional mutual information is between 0 and 1, 0 means independent variables and no correlation; if the interaction information relationship is strong, it tends to 1.

                      (2)                                  

(2) Find the variable with the maximum interaction information for each variable and connect it with an undirected arc.

(3) Transform an undirected arc into a directed arc.

(4) Output variables are connected to all input variables.

In the parameter learning of Bayesian method, the variables of each node are binary variables. Each parameter  in the parameter set of nodes is the probability of "success", and the prior probability distribution of parameters  should choose the binomial conjugate distribution. Using sample data to modify prior probability, the final value  is the expectation of parameter  posterior distribution.

 

 

 

 

 

References

[1] Talbot R., Fagerlind H., Morris, A., (2013). Exploring inattention and distraction in the SafetyNet Crash Causation Database. Accident Analysis and Prevention, 60, 445–455.

[2] Xiong X.X., Wang M., Cai Y.F., Chen L., Farah H., Hagenzieker M., (2019). A forward collision avoidance algorithm based on driver braking behavior. Accident Analysis and Prevention, 129, 30–43.

[3] Yang L.P., Feng Z.X., Zhao X.H., Jiang K., Huang Z.P., (2020). Analysis of the factors affecting drivers’ queue-jumping behaviors in China. Transportation Research Part F, 72, 96–109.

[4] Zhang Z., Guo Y.S., Fu R., Yuan W., Wang C., (2020). Linking executive functions to distracted driving, does it differ between young and mature drivers? Plos One, 15(9), e0239596.

 

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The author already modified the paper according to my comments. The version 2 significantly better than the first one. No more other questions.

The research method and objective of the following reference related with the topic of this paper, so I suggest the author quote it in the literature review.

Liu L, Ye X, Wang T, Yan X, Chen J, Ran B. Key Factors Analysis of Severity of Automobile to Two-Wheeler Traffic Accidents Based on Bayesian Network. International Journal of Environmental Research and Public Health. 2022; 19(10):6013. https://doi.org/10.3390/ijerph19106013

Author Response

Dear editor,

I uploaded the wrong version of the revision before, which brought you inconvenience. Your opinions are very useful for the improvement of my paper. This is the modification explanation of my paper. Please check it. 

Author Response File: Author Response.docx

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.