Method for cross-modal correlation analysis of traffic accident road network events based on large model

Abstract

The application discloses a traffic accident road network event cross-modal correlation analysis method based on a large model, relates to the field of public traffic, and solves the problem that the existing traffic accident road network event cross-modal correlation analysis method has poor analysis effect, and comprises the following steps: S1, moving speed analysis is performed on a participant corresponding to a target traffic accident, a target accident time period is obtained according to an analysis result, S2, a three-dimensional model is established for a target accident space region in the target accident time period, accident damage assessment is performed on the target accident space region according to an accident region three-dimensional interaction model, and accident damage assessment data is obtained, and S3, congestion duration of the target accident space region is predicted according to the accident damage assessment data, and road network shunting early warning is performed according to a prediction result, and the application can improve the accuracy and timeliness of traffic accident bypass early warning.

Description

Technical Field

[0001] This invention belongs to the field of public transportation and involves large-scale model technology, specifically a method for cross-modal correlation analysis of traffic accident road network events based on large-scale models. Background Technology

[0002] 1. Existing cross-modal correlation analysis methods for road network events, when conducting damage analysis of traffic accidents, rely on manual inspections to inventory the damage at the accident site. They do not collect data on the accident cycle based on changes in the relative movement speed of the accident participants, nor do they create a 3D interactive model of the accident area during the accident cycle. They cannot collect the intrusion volume and intrusion speed of the accident participants into the road body in real time based on the 3D interactive model of the accident area, thus failing to assess the damage to the target accident space area and resulting in a lag in the traffic accident assessment process.

[0003] 2. Existing cross-modal correlation analysis methods for road network events typically provide detour warnings for regional road networks based on the actual processing progress of traffic accidents. They fail to consistently match the degree of accident damage and road network speed corresponding to the target traffic accident with the historical traffic accidents in the region. Consequently, they cannot predict the congestion duration of the target traffic accident based on the actual congestion duration of the matched historical accidents, resulting in a lack of timeliness and accuracy in detour warnings.

[0004] To address this, we propose a cross-modal correlation analysis method for traffic accident road network events based on a large model. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the purpose of this invention is to provide a method for cross-modal correlation analysis of traffic accident road network events based on a large model. This invention aims to improve the accuracy and timeliness of the cross-modal correlation analysis method for traffic accident road network events.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a cross-modal correlation analysis method for traffic accident road network events based on a large model, comprising the following steps:

[0007] Step S1: Obtain the target traffic accident, analyze the movement speed of the participating entities corresponding to the target traffic accident, and obtain the time period of the target accident based on the analysis results;

[0008] Step S2: Perform 3D modeling of the target accident spatial area within the target accident time period to obtain a 3D interactive model of the accident area. Based on the 3D interactive model of the accident area, conduct an accident damage assessment of the target accident spatial area and obtain accident damage assessment data based on the assessment results.

[0009] Step S3: Based on the accident damage assessment data, predict the congestion duration of the target accident area and issue a road network diversion warning based on the prediction results.

[0010] Furthermore, in step S1, the specific steps are as follows:

[0011] Step S11: Obtain the traffic accident that occurred at the current moment, get the target traffic accident, and set the spatial region corresponding to the target traffic accident as the target accident spatial region;

[0012] Step S12: Analyze the movement speed of the parties involved in the target traffic accident and obtain the start time of the accident based on the analysis results;

[0013] Step S13: Obtain the accident subjects involved in the target traffic accident, and obtain multiple traffic accident subjects. Obtain the time point when each traffic accident subject is in a stationary state, and obtain the subject's movement stop time point. Sort the multiple subject movement stop time points obtained in chronological order, and set the subject movement stop time point at the last position as the accident end time point.

[0014] Step S14: Set the time interval between the start time of the accident and the end time of the accident as the target accident time period.

[0015] Furthermore, in step S12, the specific steps are as follows:

[0016] The initial accident subjects are obtained by identifying the parties who were first involved in the collision in the target traffic accident.

[0017] The time point when the initial accident subject enters the target accident space area is set as the first accident characteristic time point, and the time point when the initial accident subject stops moving in the target accident space area is set as the second accident characteristic time point. The time period between the first accident characteristic time point and the second accident characteristic time point is set as the trajectory monitoring period.

[0018] Create a speed-time coordinate system, collect real-time vehicle speed data for the initial accident subject during the trajectory monitoring period, and plot the collected results as a first-time speed curve;

[0019] The traffic accident subject that first collides with the initial accident subject is obtained to obtain the initial traffic accident subject. The real-time movement speed of the initial traffic accident subject during the trajectory monitoring period is collected, and the collection results are plotted as a second time-velocity curve.

[0020] In the velocity-time coordinate system, the coordinate range covered by the trajectory monitoring period on the x-axis is set as the time period coordinate range. A velocity monitoring window is set, and the coordinate position of the velocity monitoring window within the time period coordinate range is set as the feature coordinate position.

[0021] Furthermore, in step S12, the specific steps are as follows:

[0022] The first characteristic velocity value is obtained by acquiring the ordinate of the intersection point of the left edge of the velocity monitoring window and the first time velocity curve. The second characteristic velocity value is obtained by acquiring the ordinate of the intersection point of the right edge of the velocity monitoring window and the first time velocity curve. The difference between the first characteristic velocity value and the second characteristic velocity value is calculated, and the absolute value of the obtained difference is taken to obtain the velocity difference of the first curve window.

[0023] The vertical coordinate of the intersection point of the left edge of the speed monitoring window and the second time speed curve is obtained to obtain the third characteristic speed value. The vertical coordinate of the intersection point of the right edge of the speed monitoring window and the second time speed curve is obtained to obtain the fourth characteristic speed value. The difference between the third characteristic speed value and the fourth characteristic speed value is calculated, and the absolute value of the obtained difference is taken to obtain the speed difference of the second curve window.

[0024] The sum of the velocity difference between the first curve window and the velocity difference between the second curve window is calculated to obtain the velocity difference of the accident subject window corresponding to the velocity monitoring window;

[0025] The speed monitoring window is used to slide through the time period coordinate range. Based on the traversal results, the speed difference of the accident main window corresponding to the speed monitoring window at different coordinate positions is obtained. The obtained speed difference of the accident main window is compared in magnitude. The speed monitoring window corresponding to the largest speed difference of the accident main window is marked as the collision time window. The median value of the time period covered by the collision time window is obtained to obtain the accident start time point.

[0026] Furthermore, in step S2, the specific steps are as follows:

[0027] Step S21: Monitor and collect images of the target accident spatial area during the target accident time period to obtain accident image video stream data; collect radar point cloud data of the target accident spatial area during the target accident time period to obtain accident area point cloud data; create a three-dimensional model of the target spatial area based on the accident image video stream data and the accident area point cloud data to obtain a three-dimensional interactive model of the accident area.

[0028] Step S22: Obtain the initial accident subject, and extract the interaction time period of the three-dimensional interactive model of the accident area where the initial accident subject is in a moving state to obtain the phased three-dimensional interactive model corresponding to the initial accident subject.

[0029] Step S23: Capture the phased 3D interactive model frame by frame to obtain multiple static phased 3D models. Set two static phased 3D models with adjacent capture times as a phased model control group to obtain multiple phased model control groups.

[0030] Step S24: Randomly select one sample model control group from the multiple stage model control groups obtained, select the feature road subject in the stage model control group, perform accident damage analysis on the initial accident subject and the feature road subject in the stage model control group, and obtain the intrusion damage degree of the initial accident subject to the feature road subject based on the analysis results.

[0031] Step S25: Obtain the degree of intrusion and damage to the feature road body for each traffic accident subject, and obtain multiple degrees of intrusion and damage;

[0032] Step S26: Obtain the cumulative intrusion volume of each traffic accident subject in the region of the characteristic road subject, and sum them to obtain the total intrusion volume of the subject;

[0033] Step S27: Set the intrusion damage degree corresponding to the same traffic accident subject to B1, the cumulative intrusion volume of the area to B2, and the total intrusion volume of the subject to B3. Calculate B1×(B2 / B3) to obtain the weighted regional damage degree. Sum the obtained multiple weighted regional damage degrees to obtain the accident damage degree corresponding to the characteristic road subject.

[0034] Step S28: Obtain the accident damage level of each road body to obtain accident damage assessment data.

[0035] Furthermore, in step S24, the specific steps are as follows:

[0036] The two static staged 3D models in the sample model control group were named the first static 3D model and the second static 3D model according to the order of the extraction time.

[0037] The road subject that comes into contact with the initial accident subject in the second static three-dimensional model is acquired, and a feature road subject is arbitrarily selected from the acquired road subjects;

[0038] In the first static 3D model, the spatial pixels occupied by the initial accident subject are set as initial spatial pixels, and the spatial pixels occupied by the feature road subject are set as feature spatial pixels. Feature spatial pixels that overlap with the initial spatial pixels are collected to obtain feature overlapping pixels, and the model space area occupied by the feature overlapping pixels is obtained to obtain the first subject intrusion area.

[0039] In the second static 3D model, the model space region occupied by the overlapping feature pixels is obtained to obtain the second main body intrusion region;

[0040] The newly added region of the second subject intrusion region relative to the first subject intrusion region is obtained to obtain the newly added intrusion region of the feature road subject in the sample model control group;

[0041] Repeat the process of obtaining the newly intruded areas of the feature road subject in the sample model control group, obtain the newly intruded areas of the feature road subject in the model control group at each stage, obtain the volume of each newly intruded area, obtain multiple intruded area volumes, sum the obtained multiple intruded area volumes, and obtain the cumulative intrusion volume of the region.

[0042] The movement speed of the initial accident subject in each sample model control group was collected to obtain the movement speed of multiple intruders.

[0043] Furthermore, in step S24, the specific steps are as follows:

[0044] Set the volume of the intrusion area corresponding to the same newly added intrusion area as A1, the cumulative intrusion volume of the area as A2, and the movement speed of the intruder as A3. Calculate A3×(A1 / A2) to obtain the weighted intrusion speed of the newly added area. Obtain the weighted intrusion speed of the newly added area corresponding to each newly added intrusion area and sum them up to obtain the weighted intrusion speed of the initial accident subject to the feature road subject. Obtain the road speed limit corresponding to the target accident space area to obtain the regional road speed limit.

[0045] The degree of intrusion damage of the initial accident subject to the characteristic road subject is obtained by calculating the weighted intrusion speed, regional road speed limit, and regional cumulative intrusion volume.

[0046] The degree of intrusion damage is calculated using the following formula:

[0047] ;

[0048] Where Phd represents the degree of intrusion and damage of the initial accident subject to the characteristic road subject, Vqr represents the weighted intrusion speed, Vzx represents the speed limit of the regional road, and Tjq represents the cumulative intrusion volume of the region.

[0049] Furthermore, step S3 also includes the following steps:

[0050] Step S31: Obtain accident damage assessment data, and obtain the accident damage degree corresponding to each road body based on the accident damage assessment data to obtain multiple target accident damage degrees;

[0051] Step S32: Obtain historical traffic accidents that occurred in the target accident space area, and screen the historical traffic accidents into a first type of matchable accident and a second type of matchable accident;

[0052] Step S33: Periodically match the first type of accident to be matched with the target traffic accident on the road network speed, and obtain multiple matching accidents based on the matching results;

[0053] Step S34: Obtain the historical congestion duration corresponding to each matching accident, calculate the average of the multiple historical congestion durations to obtain the first duration prediction feature value, calculate the standard deviation of the multiple historical congestion durations to obtain the second duration prediction feature value, and calculate the sum of the first duration prediction feature value and the second duration prediction feature value to obtain the predicted congestion duration corresponding to the target traffic accident.

[0054] Step S35: Set a reasonable range for congestion duration. If the predicted congestion duration is within the reasonable range, there is no need to issue a detour warning for the road network analysis area. If the predicted congestion duration is not within the reasonable range, a detour warning will be issued for the road network analysis area.

[0055] Furthermore, step S32 also includes the following steps:

[0056] Randomly select a sample historical traffic accident from the acquired historical traffic accidents, obtain the accident damage degree corresponding to each road body of the sample historical accident, and name it as the historical accident damage degree.

[0057] For the same road body, the damage degree of the target accident is set as C1 and the damage degree of the historical accident is set as C2. The damage deviation ratio corresponding to the road body is obtained by calculating |C1-C2| / C1. The damage deviation ratio corresponding to each road body is obtained and summed to obtain the accident damage deviation degree between the sample historical accident and the target traffic accident.

[0058] Obtain the accident damage deviation between each historical traffic accident and the target traffic accident, and set the accident damage deviation benchmark range;

[0059] If the accident damage deviation is within the accident damage deviation benchmark range, the corresponding historical traffic accident will be classified as a first type of accident to be matched. If the accident damage deviation is not within the accident damage deviation benchmark range, the corresponding historical traffic accident will be classified as a second type of accident to be matched.

[0060] Furthermore, step S33 also includes the following steps:

[0061] A road network analysis area is set up outside the target accident space area. Each road within the road network analysis area is acquired, and the average vehicle speed of each road during the target accident time period is obtained to obtain multiple target vehicle speeds.

[0062] Select a sample accident to be matched from the first type of accident to be matched, and obtain the average vehicle speed of each road within the accident time period corresponding to the sample accident to be matched, so as to obtain multiple sample vehicle speeds.

[0063] For the same road, the target vehicle's speed is set to D1 and the sample vehicle's speed is set to D2. The speed deviation ratio corresponding to the road is calculated by |D1-D2| / D1. The speed deviation ratio corresponding to each road is obtained and summed to obtain the road network vehicle speed deviation between the sample accident to be matched and the target traffic accident.

[0064] Obtain the road network vehicle speed deviation between each first-type matchable accident and the target traffic accident, and set the vehicle speed deviation benchmark range;

[0065] If the road network speed deviation is within the speed deviation benchmark range, the corresponding first type of accident to be matched will be classified as a matched accident. If the road network speed deviation is not within the speed deviation benchmark range, the corresponding first type of accident to be matched will be classified as a non-matched accident.

[0066] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0067] 1. This invention collects accident cycle data based on the change in the relative movement speed of the accident participants, creates a three-dimensional interactive model of the accident area in the accident cycle, and collects the intrusion volume and intrusion speed of the accident participants on the road in real time based on the three-dimensional interactive model of the accident area, and conducts accident damage assessment on the target accident space area accordingly, thereby improving the efficiency of traffic accident damage assessment.

[0068] 2. This invention matches the degree of damage to the target traffic accident and the road network speed with the historical traffic accidents in the area, and predicts the congestion duration of the target traffic accident based on the actual congestion duration of the matched historical accidents, thereby improving the timeliness and accuracy of detour warnings. Attached Figure Description

[0069] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0070] Figure 1 This is a diagram illustrating the implementation steps of the present invention;

[0071] Figure 2 This is the velocity-time coordinate system in this invention;

[0072] Figure 3 This is a schematic diagram of the curve intersection points of the present invention. Detailed Implementation

[0073] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0074] Example 1

[0075] Please see Figure 1 This invention provides a technical solution: a method for cross-modal correlation analysis of traffic accident road network events based on a large model, comprising the following steps:

[0076] Step S1: Obtain the target traffic accident, analyze the movement speed of the participating entities corresponding to the target traffic accident, and obtain the time period of the target accident based on the analysis results;

[0077] In step S1, the specific steps are as follows:

[0078] The system acquires information on traffic accidents occurring at the current moment, obtains the target traffic accident, and sets the spatial region corresponding to the target traffic accident as the target accident spatial region.

[0079] It should be noted here that:

[0080] In this application, the target accident spatial region referred to herein is the spatial region where cross-modal correlation analysis of traffic accident road network events is deployed;

[0081] Analyze the movement speed of the parties involved in the target traffic accident, and obtain the time period of the target accident based on the analysis results;

[0082] Specifically as follows:

[0083] The initial accident subjects are obtained by identifying the parties who were first involved in the collision in the target traffic accident.

[0084] The time point when the initial accident subject enters the target accident space area is set as the first accident characteristic time point, and the time point when the initial accident subject stops moving in the target accident space area is set as the second accident characteristic time point. The time period between the first accident characteristic time point and the second accident characteristic time point is set as the trajectory monitoring period.

[0085] Please see Figure 2A speed-time coordinate system is created to collect real-time vehicle speed data for the initial accident subject during the trajectory monitoring period, and the collected results are plotted as the first-time speed curve.

[0086] It should be noted here that:

[0087] In this application, the x-axis of the velocity-time coordinate system represents time, and the y-axis represents velocity.

[0088] The traffic accident subject that first collides with the initial accident subject is obtained to obtain the initial traffic accident subject. The real-time movement speed of the initial traffic accident subject during the trajectory monitoring period is collected, and the collection results are plotted as a second time-velocity curve.

[0089] It should be noted here that:

[0090] In this application, the subject of the initial traffic accident is specifically any object appearing in the target accident space area, including but not limited to pedestrians, vehicles, and road guardrails.

[0091] In the velocity-time coordinate system, the coordinate range covered by the x-axis during the trajectory monitoring period is set as the time period coordinate range. A velocity monitoring window is set, and the coordinate position of the velocity monitoring window within the time period coordinate range is set as the feature coordinate position.

[0092] It should be noted here that:

[0093] In this application, the velocity monitoring window is a time interval used to dynamically capture local velocity features, and the interval length corresponding to the velocity monitoring window is 1 second.

[0094] The first characteristic velocity value is obtained by acquiring the ordinate of the intersection point of the left edge of the velocity monitoring window and the first time velocity curve. The second characteristic velocity value is obtained by acquiring the ordinate of the intersection point of the right edge of the velocity monitoring window and the first time velocity curve. The difference between the first characteristic velocity value and the second characteristic velocity value is calculated, and the absolute value of the obtained difference is taken to obtain the velocity difference of the first curve window.

[0095] It should be noted here that:

[0096] Please see Figure 3 In this application, the intersection of the left edge of the speed monitoring window with the first time speed curve is the first curve intersection point, the intersection of the right edge of the speed monitoring window with the first time speed curve is the first curve intersection point, the intersection of the left edge of the speed monitoring window with the second time speed curve is the first curve intersection point, and the intersection of the right edge of the speed monitoring window with the second time speed curve is the first curve intersection point.

[0097] The vertical coordinate of the intersection point of the left edge of the speed monitoring window and the second time speed curve is obtained to obtain the third characteristic speed value. The vertical coordinate of the intersection point of the right edge of the speed monitoring window and the second time speed curve is obtained to obtain the fourth characteristic speed value. The difference between the third characteristic speed value and the fourth characteristic speed value is calculated, and the absolute value of the obtained difference is taken to obtain the speed difference of the second curve window.

[0098] The sum of the velocity difference between the first curve window and the velocity difference between the second curve window is calculated to obtain the velocity difference of the accident subject window corresponding to the velocity monitoring window;

[0099] The speed monitoring window is used to slide through the time period coordinate range. Based on the traversal results, the speed difference of the accident main window corresponding to the speed monitoring window at different coordinate positions is obtained. The obtained speed difference of the accident main window is compared. The speed monitoring window corresponding to the largest speed difference of the accident main window is marked as the collision time window. The median value of the time period covered by the collision time window is obtained to obtain the accident start time point.

[0100] The parties involved in the target traffic accident are obtained separately, resulting in multiple traffic accident parties. The time point when each traffic accident party is in a stationary state is obtained, resulting in the time point when the party stops moving. The multiple time points when the parties stop moving are sorted in chronological order, and the time point when the party that stops moving at the end of the list is set as the time point when the accident ends.

[0101] It should be noted here that:

[0102] In this application, the subject of the traffic accident is specifically a vehicle, including but not limited to vans, cars, and motorcycles.

[0103] Set the time interval between the start time of the accident and the end time of the accident as the target accident time period;

[0104] Step S2: Perform 3D modeling of the target accident spatial area within the target accident time period to obtain a 3D interactive model of the accident area. Based on the 3D interactive model of the accident area, conduct an accident damage assessment of the target accident spatial area and obtain accident damage assessment data based on the assessment results.

[0105] Specifically as follows:

[0106] Monitoring images of the target accident spatial area during the target accident time period are acquired to obtain accident image video stream data. Radar point cloud data of the target accident spatial area during the target accident time period is acquired to obtain accident area point cloud data. Based on the accident image video stream data and the accident area point cloud data, a three-dimensional model of the target spatial area is created to obtain a three-dimensional interactive model of the accident area.

[0107] It should be noted here that:

[0108] In this application, the existing algorithm used to create the three-dimensional interactive model of the accident area is a deep learning-based image three-dimensional reconstruction algorithm;

[0109] It should be noted here that:

[0110] In this application, the three-dimensional interactive model of the accident area involved herein is specifically a three-dimensional display of video stream images.

[0111] The initial accident subject is obtained, and the interaction time period is extracted from the 3D interactive model of the accident area where the initial accident subject is in a moving state to obtain the staged 3D interactive model corresponding to the initial accident subject.

[0112] The phased 3D interactive model is frame-by-frame to obtain multiple static phased 3D models. Two static phased 3D models with adjacent frame extraction times are set as a phased model control group to obtain multiple phased model control groups.

[0113] Randomly select one sample model control group from the multiple stage model control groups obtained, and name the two static stage 3D models in the sample model control group as the first static 3D model and the second static 3D model according to the order of the extraction time.

[0114] It should be noted here that:

[0115] In this application, the model capture time corresponding to the first static 3D model is earlier than the model capture time corresponding to the second static 3D model.

[0116] The road subject that comes into contact with the initial accident subject in the second static three-dimensional model is acquired, and a feature road subject is arbitrarily selected from the acquired road subjects;

[0117] It should be noted here that:

[0118] In this application, the road entities referred to herein include, but are not limited to, road surfaces, guardrails, and vehicles.

[0119] In the first static 3D model, the spatial pixels occupied by the initial accident subject are set as initial spatial pixels, and the spatial pixels occupied by the feature road subject are set as feature spatial pixels. Feature spatial pixels that overlap with the initial spatial pixels are collected to obtain feature overlapping pixels, and the model space area occupied by the feature overlapping pixels is obtained to obtain the first subject intrusion area.

[0120] In the second static 3D model, the model space region occupied by the overlapping feature pixels is obtained to obtain the second main body intrusion region;

[0121] The newly added region of the second subject intrusion region relative to the first subject intrusion region is obtained to obtain the newly added intrusion region of the feature road subject in the sample model control group;

[0122] Repeat the process of obtaining the newly intruded areas of the feature road subject in the sample model control group, obtain the newly intruded areas of the feature road subject in the model control group at each stage, obtain the volume of each newly intruded area, obtain multiple intruded area volumes, sum the obtained multiple intruded area volumes, and obtain the cumulative intrusion volume of the region.

[0123] The movement speed of the initial accident subject in each sample model control group was collected to obtain the movement speed of multiple intruders;

[0124] Set the volume of the intrusion area corresponding to the same newly added intrusion area as A1, the cumulative intrusion volume of the area as A2, and the movement speed of the intruder as A3. Calculate A3×(A1 / A2) to obtain the weighted intrusion speed of the newly added area. Obtain the weighted intrusion speed of the newly added area corresponding to each newly added intrusion area and sum them up to obtain the weighted intrusion speed of the initial accident subject to the feature road subject. Obtain the road speed limit corresponding to the target accident space area to obtain the regional road speed limit.

[0125] The degree of intrusion damage of the initial accident subject to the characteristic road subject is obtained by calculating the weighted intrusion speed, regional road speed limit, and regional cumulative intrusion volume.

[0126] The degree of intrusion damage is calculated using the following formula:

[0127] ;

[0128] Where Phd represents the degree of intrusion and damage of the initial accident subject to the characteristic road subject, Vqr represents the weighted intrusion speed, Vzx represents the speed limit of the regional road, and Tjq represents the cumulative intrusion volume of the region;

[0129] It should be noted here that:

[0130] In this application, the ratio of weighted intrusion speed to regional road speed limit reflects the relative magnitude of the intrusion speed with respect to the road speed limit. If the intrusion speed is close to or exceeds the road speed limit, the ratio will be larger, indicating that the intrusion event is more serious. The ratio multiplied by the cumulative intrusion volume of the region represents the range of physical impact of the intrusion event on the main road. The larger the volume, the greater the degree of damage. Multiplying the speed ratio by the intrusion volume can comprehensively consider the speed and scale of the intrusion event, thereby more comprehensively assessing the degree of damage caused by traffic accidents.

[0131] In practical applications, the following test data exists:

[0132] Test data 1: Vqr=50km / h, Vzx=60km / h, Tjq=10m 3 Phd≈8.33;

[0133] Test data 2: Vqr=80km / h, Vzx=60km / h, Tjq=5m 3 Phd≈6.67;

[0134] Test data 3: Vqr=70km / h, Vzx=70km / h, Tjq=20m 3 Phd≈20.

[0135] Repeat the initial accident subject's intrusion and damage degree on the characteristic road subject, and obtain the intrusion and damage degree of each traffic accident subject on the characteristic road subject, resulting in multiple intrusion and damage degrees;

[0136] The cumulative intrusion volume of each traffic accident subject to the characteristic road subject is obtained separately, and then summed to obtain the total intrusion volume of the subject.

[0137] Let B1 be the degree of intrusion damage corresponding to the same traffic accident subject, B2 be the cumulative intrusion volume of the area, and B3 be the total intrusion volume of the subject. Calculate B1×(B2 / B3) to obtain the weighted regional damage degree. Sum the multiple weighted regional damage degrees to obtain the accident damage degree corresponding to the characteristic road subject.

[0138] Repeat the process of assessing the accident damage degree corresponding to the main feature road body, and obtain the accident damage degree for each main road body to obtain accident damage assessment data;

[0139] It should be noted here that:

[0140] In this application, the road subject involved here includes all objects that may exist in the target accident space area, and there are objects with an accident damage degree of 0, that is, objects that have not been damaged by the accident.

[0141] It should be noted here that:

[0142] Step S2 above precisely defines the accident cycle range by dynamically capturing changes in the relative movement speeds of the accident participants and constructing a three-dimensional interactive model of the accident area. This model can simulate the spatial interaction between the accident participants and the road environment in real time, and quantitatively analyze the intrusion volume and speed of the accident participants into road facilities. This dynamic and three-dimensional assessment method breaks through the limitations of traditional two-dimensional planar analysis, making the assessment of the degree of accident damage closer to the actual scenario. By collecting intrusion data in real time, the system can quickly generate a damage assessment report, significantly improving assessment efficiency, providing a scientific basis for subsequent emergency response and road repair, and reducing subjective errors from human judgment.

[0143] Step S3: Based on the accident damage assessment data, predict the congestion duration of the target accident area and issue a road network diversion warning based on the prediction results;

[0144] Step S3 further includes the following steps:

[0145] Obtain accident damage assessment data, and based on the accident damage assessment data, obtain the accident damage degree corresponding to each main road body to obtain multiple target accident damage degrees;

[0146] Historical traffic accidents occurring in the target accident spatial area are acquired, and a sample historical accident is randomly selected from the acquired historical traffic accidents.

[0147] Obtain the accident damage level corresponding to each road body in the sample historical accidents, and name it as the historical accident damage level;

[0148] For the same road body, the damage degree of the target accident is set as C1 and the damage degree of the historical accident is set as C2. The damage deviation ratio corresponding to the road body is obtained by calculating |C1-C2| / C1. The damage deviation ratio corresponding to each road body is obtained and summed to obtain the accident damage deviation degree between the sample historical accident and the target traffic accident.

[0149] Repeat the comparison of accident damage deviation between the sample historical accidents and the target traffic accident, obtain the accident damage deviation between each historical traffic accident and the target traffic accident, and set the accident damage deviation benchmark range.

[0150] If the accident damage deviation is within the accident damage deviation benchmark range, the corresponding historical traffic accident will be classified as a first type of accident to be matched; if the accident damage deviation is not within the accident damage deviation benchmark range, the corresponding historical traffic accident will be classified as a second type of accident to be matched.

[0151] It should be noted here that:

[0152] In this application, the accident damage deviation benchmark interval is 0, that is, there is no damage deviation between two traffic accidents. Multiple first-type matching accidents corresponding to the historical traffic accidents that have completed damage degree matching are obtained, and the accident damage deviation degree between each first-type matching accident and the corresponding historical traffic accident is obtained. The accident damage deviation degree with the largest value is set as the upper limit of the accident damage deviation benchmark interval.

[0153] In this application, the first type of accident to be matched includes cases where the accident damage deviation is at the boundary of the accident damage deviation reference range.

[0154] A road network analysis area is set up outside the target accident space area. Each road within the road network analysis area is acquired, and the average vehicle speed of each road during the target accident time period is obtained to obtain multiple target vehicle speeds.

[0155] Select a sample accident to be matched from the first type of accident to be matched, and obtain the average vehicle speed of each road within the accident time period corresponding to the sample accident to be matched, so as to obtain multiple sample vehicle speeds.

[0156] For the same road, the target vehicle's speed is set to D1 and the sample vehicle's speed is set to D2. The speed deviation ratio corresponding to the road is calculated by |D1-D2| / D1. The speed deviation ratio corresponding to each road is obtained and summed to obtain the road network vehicle speed deviation between the sample accident to be matched and the target traffic accident.

[0157] The process of obtaining the road network vehicle speed deviation between repeated sample matching accidents and target traffic accidents involves obtaining the road network vehicle speed deviation between each first type of matching accident and target traffic accident, and setting the vehicle speed deviation benchmark range.

[0158] It should be noted here that:

[0159] In this application, the lower limit of the benchmark interval for vehicle speed deviation is 0, meaning that there is no destructive deviation between the two traffic accidents. Multiple matching accidents corresponding to historical traffic accidents that match the vehicle speed deviation of the road network are obtained. The road network vehicle speed deviation of each matching accident and the corresponding historical traffic accident is obtained, and the road network vehicle speed deviation with the largest value is set as the upper limit of the benchmark interval for vehicle speed deviation.

[0160] If the road network speed deviation is within the speed deviation benchmark range, the corresponding first type of accident to be matched will be classified as a matched accident; if the road network speed deviation is not within the speed deviation benchmark range, the corresponding first type of accident to be matched will be classified as a non-matched accident.

[0161] Obtain the historical congestion duration corresponding to each matching accident, calculate the average of the obtained historical congestion durations to obtain the first duration prediction feature value, calculate the standard deviation of the obtained historical congestion durations to obtain the second duration prediction feature value, and calculate the sum of the first duration prediction feature value and the second duration prediction feature value to obtain the predicted congestion duration corresponding to the target traffic accident.

[0162] It should be noted here that:

[0163] In this application, the historical congestion durations mentioned herein can be collected through a traffic information sharing platform.

[0164] In this application, the average value is calculated as the first duration prediction feature value by analyzing the congestion duration of matching accidents in historical data, which represents the average level of historical congestion duration.

[0165] Next, the standard deviation is calculated as the second duration prediction feature value to measure the dispersion of historical data and the uncertainty of future congestion duration. The sum of these two feature values ​​yields a predicted congestion duration that not only covers typical cases but also reserves redundancy for potentially longer congestion. This helps traffic management departments to more comprehensively assess congestion risks, issue early warnings of detours for passing vehicles, and ensure smooth and safe traffic.

[0166] Set a reasonable range for congestion duration. If the predicted congestion duration is within the reasonable range, there is no need to issue a detour warning for the road network analysis area. If the predicted congestion duration is not within the reasonable range, a detour warning will be issued for the road network analysis area.

[0167] It should be noted here that:

[0168] In this application, it is not necessary to issue detour warnings for the road network analysis area, including situations where the predicted congestion duration falls within the boundary of a reasonable congestion duration range;

[0169] In this application, the reasonable range for congestion duration is 0, that is, no traffic congestion occurs. Historical traffic accidents that issued detour warnings are obtained, and the actual congestion duration corresponding to each historical traffic accident is obtained. The actual congestion duration with the smallest value is set as the upper limit of the reasonable range for congestion duration.

[0170] It should be noted here that:

[0171] Step S3 above matches the damage level of the target accident with the road network speed characteristics to a historical accident database, selecting the actual congestion duration under similar scenarios as the prediction benchmark. This method fully utilizes the accumulated value of historical data, avoiding the limitations of relying solely on theoretical models or real-time data. By matching the congestion duration of consistent historical accidents, the prediction results are closer to reality, effectively improving the timeliness and accuracy of detour warnings. Simultaneously, this mechanism can dynamically adapt to the congestion characteristics of different regions and different types of accidents, providing traffic management departments with more targeted early warning information, thereby optimizing traffic management strategies and reducing the risk of secondary congestion.

[0172] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for cross-modal correlation analysis of traffic accident road network events based on a large model, characterized in that, include: Step S1: Obtain the target traffic accident, analyze the movement speed of the participating entities in the target traffic accident, and obtain the time period of the target accident based on the analysis results; Step S2: Perform 3D modeling of the target accident spatial area within the target accident time period to obtain a 3D interactive model of the accident area. Based on the 3D interactive model of the accident area, conduct an accident damage assessment of the target accident spatial area and obtain accident damage assessment data based on the assessment results. In step S2, the specific steps are as follows: Step S21: Monitor and acquire images of the target accident space area to obtain accident image video stream data; acquire radar point cloud data of the target accident space area to obtain accident area point cloud data. A 3D model was created to obtain a 3D interactive model of the accident area. Step S22: Obtain the initial accident subject, and extract the interaction time period of the three-dimensional interactive model of the accident area where the initial accident subject is in a moving state to obtain a phased three-dimensional interactive model; Step S23: Capture the phased 3D interactive model frame by frame to obtain multiple static phased 3D models. Set any two static phased 3D models with adjacent capture times as a phased model control group to obtain multiple phased model control groups. Step S24: Select the main feature road in the stage model control group, and perform accident damage analysis on the initial accident subject and the main feature road subject in the stage model control group to obtain the degree of intrusion damage of the initial accident subject to the main feature road subject; The specific steps in step S24 are as follows: Select a sample model control group, and name the two static stage 3D models in the sample model control group as the first static 3D model and the second static 3D model according to the order of the extraction time. Arbitrarily select a characteristic road subject from the road subjects that come into contact with the initial accident subject in the second static three-dimensional model; In the first static 3D model, the spatial pixels occupied by the initial accident subject are set as initial spatial pixels, and the spatial pixels occupied by the feature road subject are set as feature spatial pixels. Feature spatial pixels that overlap with the initial spatial pixels are collected to obtain feature overlapping pixels, and the model space area occupied by the feature overlapping pixels is obtained to obtain the first subject intrusion area. In the second static 3D model, the model space region occupied by the overlapping feature pixels is obtained to obtain the second main body intrusion region; The newly added region of the second subject intrusion region relative to the first subject intrusion region is obtained to obtain the newly added intrusion region of the feature road subject in the sample model control group; The newly intrusive regions of the main feature road body in the model control group at each stage are obtained. The volume of each newly intrusive region is obtained, and the volumes of multiple intrusive regions are summed to obtain the cumulative intrusive volume of the region. The movement speed of the initial accident subject in each sample model control group was collected to obtain the movement speed of multiple intruders; Set the volume of the intrusion area corresponding to the same newly added intrusion area as A1, the cumulative intrusion volume of the area as A2, and the movement speed of the intruder as A3. Calculate A3×(A1 / A2) to obtain the weighted intrusion speed of the newly added area. Summate the weighted intrusion speed of the newly added area corresponding to each newly added intrusion area to obtain the weighted intrusion speed of the initial accident subject to the feature road subject. Obtain the road speed limit corresponding to the target accident space area to obtain the regional road speed limit. The degree of intrusion damage to the characteristic road body by the initial accident subject is obtained by calculating the weighted intrusion speed, regional road speed limit, and regional cumulative intrusion volume. Step S25: Obtain the degree of intrusion and damage of each traffic accident subject to the feature road subject, and obtain multiple degrees of intrusion and damage; Step S26: Obtain the cumulative intrusion volume of each traffic accident subject in the region of the characteristic road subject, and sum them to obtain the total intrusion volume of the subject; Step S27: Set the intrusion damage degree corresponding to the same traffic accident subject to B1, the cumulative intrusion volume of the area to B2, and the total intrusion volume of the subject to B3. Calculate the weighted regional damage degree, sum the obtained multiple weighted regional damage degrees, and obtain the accident damage degree corresponding to the characteristic road subject. Step S28: Obtain the degree of accident damage corresponding to each main road structure to obtain accident damage assessment data; Step S3: Based on the accident damage assessment data, predict the congestion duration of the target accident area and issue a road network diversion warning based on the prediction results; Step S3 further includes the following steps: Step S31: Obtain accident damage assessment data, and obtain the accident damage degree corresponding to each road body based on the accident damage assessment data to obtain multiple target accident damage degrees; Step S32: Obtain historical traffic accidents that occurred in the target accident space area, and screen the historical traffic accidents into a first type of matchable accident and a second type of matchable accident; Step S33: Periodically match the first type of accident to be matched with the target traffic accident on the road network speed, and obtain multiple matching accidents based on the matching results; Step S34: Obtain the historical congestion duration corresponding to each matching accident, calculate the mean of the multiple historical congestion durations to obtain the first duration prediction feature value, calculate the standard deviation of the multiple historical congestion durations to obtain the second duration prediction feature value, and calculate the sum of the first duration prediction feature value and the second duration prediction feature value to obtain the predicted congestion duration corresponding to the target traffic accident. Step S35: Set a reasonable range for congestion duration. If the predicted congestion duration is within the reasonable range, there is no need to issue a detour warning for the road network analysis area. If it is not within the reasonable range, then issue a detour warning for the road network analysis area.

2. The method for cross-modal correlation analysis of traffic accident road network events based on a large model according to claim 1, characterized in that, In step S1, the specific steps are as follows: Step S11: Obtain the traffic accident that occurred at the current moment, get the target traffic accident, and set the spatial region corresponding to the target traffic accident as the target accident spatial region; Step S12: Obtain the initial accident subject and perform movement speed analysis, and obtain the accident start time point based on the analysis results; Step S13: Obtain the time point when each traffic accident subject is stationary to obtain the subject's movement stop time point. Sort the multiple subject movement stop time points obtained in chronological order and set the subject movement stop time point at the end of the list as the accident end time point. Step S14: Set the time interval between the start time of the accident and the end time of the accident as the target accident time period.

3. The method for cross-modal correlation analysis of traffic accident road network events based on a large model according to claim 2, characterized in that, In step S12, the specific steps are as follows: The initial accident subjects are obtained by identifying the parties who were first involved in the collision in the target traffic accident. The time point when the initial accident subject enters the target accident space area is set as the first accident characteristic time point, and the time point when the initial accident subject stops moving in the target accident space area is set as the second accident characteristic time point. The time period between the first accident characteristic time point and the second accident characteristic time point is set as the trajectory monitoring period. Create a speed-time coordinate system, collect real-time vehicle speed data for the initial accident subject during the trajectory monitoring period, and plot the first time-time speed curve; The traffic accident subject that first collides with the initial accident subject is obtained to obtain the initial traffic accident subject. The real-time movement speed of the initial traffic accident subject during the trajectory monitoring period is collected, and a second time-velocity curve is plotted. In the velocity-time coordinate system, the coordinate range covered by the x-axis during the trajectory monitoring period is set as the time period coordinate range. A velocity monitoring window is set, and the coordinate position of the velocity monitoring window within the time period coordinate range is set as the feature coordinate position.

4. The method for cross-modal correlation analysis of traffic accident road network events based on a large model according to claim 3, characterized in that, In step S12, the specific steps are as follows: The first characteristic velocity value is obtained by acquiring the ordinate of the intersection point of the left edge of the velocity monitoring window and the first time velocity curve. The second characteristic velocity value is obtained by acquiring the ordinate of the intersection point of the right edge of the velocity monitoring window and the first time velocity curve. The difference between the first characteristic velocity value and the second characteristic velocity value is calculated, and the absolute value of the obtained difference is taken to obtain the velocity difference of the first curve window. The vertical coordinate of the intersection point of the left edge of the speed monitoring window and the second time speed curve is obtained to obtain the third characteristic speed value. The vertical coordinate of the intersection point of the right edge of the speed monitoring window and the second time speed curve is obtained to obtain the fourth characteristic speed value. The difference between the third characteristic speed value and the fourth characteristic speed value is calculated, and the absolute value of the obtained difference is taken to obtain the speed difference of the second curve window. The sum of the velocity difference between the first curve window and the velocity difference between the second curve window is calculated to obtain the velocity difference of the accident subject window corresponding to the velocity monitoring window; The speed monitoring window is used to slide through the time period coordinate range. Based on the traversal results, the speed difference of the accident main window corresponding to the speed monitoring window at different coordinate positions is obtained. The obtained speed difference of the accident main window is compared in magnitude. The speed monitoring window corresponding to the largest speed difference of the accident main window is marked as the collision time window. The median value of the time period covered by the collision time window is obtained to obtain the accident start time point.

5. The method for cross-modal correlation analysis of traffic accident road network events based on a large model according to claim 1, characterized in that, Step S32 further includes the following steps: Randomly select a sample historical traffic accident from the acquired historical traffic accidents, obtain the accident damage degree corresponding to each road body of the sample historical accident, and name it as the historical accident damage degree. For the same road body, the damage degree of the target accident is set as C1 and the damage degree of the historical accident is set as C2. Calculate |C1-C2| / C1 to obtain the damage deviation ratio corresponding to the road body. Obtain the damage deviation ratio corresponding to each road body and sum them to obtain the accident damage deviation degree between the sample historical accident and the target traffic accident. Obtain the accident damage deviation between each historical traffic accident and the target traffic accident, and set the accident damage deviation benchmark range; If the accident damage deviation is within the accident damage deviation benchmark range, the corresponding historical traffic accident will be classified as a first type of accident to be matched; otherwise, the corresponding historical traffic accident will be classified as a second type of accident to be matched.

6. The method for cross-modal correlation analysis of traffic accident road network events based on a large model according to claim 1, characterized in that, Step S33 further includes the following steps: A road network analysis area is set up outside the target accident space area. The average vehicle speed of each road within the road network analysis area is obtained, and the average vehicle speed of each road within the target accident time period is obtained, thus obtaining multiple target vehicle speeds. Select a sample accident to be matched from the first type of accident to be matched, and obtain the average vehicle speed of each road within the accident time period corresponding to the sample accident to be matched, so as to obtain multiple sample vehicle speeds. For the same road, the target vehicle's speed is set to D1 and the sample vehicle's speed is set to D2. The speed deviation ratio corresponding to the road is calculated by |D1-D2| / D1. The speed deviation ratio corresponding to each road is obtained and summed to obtain the road network vehicle speed deviation between the sample accident to be matched and the target traffic accident. Obtain the road network vehicle speed deviation between each first-type accident to be matched and the target traffic accident, and set the vehicle speed deviation benchmark range; If the road network speed deviation is within the speed deviation benchmark range, the corresponding first type of accident to be matched will be classified as a matched accident; otherwise, the corresponding first type of accident to be matched will be classified as a non-matched accident.

Images