CN-121438560-B - Traffic accident road network event cross-modal correlation analysis method based on large model

Abstract

The invention discloses a traffic accident road network event cross-modal correlation analysis method based on a large model, relates to the field of public transportation, 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 of S1: the method comprises the steps of S2, carrying out three-dimensional modeling on a target accident space region in the target accident time period according to analysis results, carrying out accident damage evaluation on the target accident space region according to an accident region three-dimensional interaction model, and obtaining accident damage evaluation data, S3, predicting congestion time of the target accident space region according to the accident damage evaluation data, and carrying out road network diversion early warning according to prediction results.

Inventors

• ZHANG WENYING
• QIU YUBO
• Meng Yayang

Assignees

• 天津市天益达科技发展有限公司

Dates

Publication Date

20260512

Application Date

20251031

Claims (6)

1. 1. The traffic accident road network event cross-modal correlation analysis method based on the large model is characterized by comprising the following steps of: Step S1, acquiring a target traffic accident, analyzing the moving speed of a participated main body of the target traffic accident, and obtaining a target accident time period according to an analysis result; Step S2, carrying out three-dimensional modeling on a target accident space region in a target accident time period to obtain an accident region three-dimensional interaction model, carrying out accident damage evaluation on the target accident space region according to the accident region three-dimensional interaction model, and obtaining accident damage evaluation data according to an evaluation result; in the step S2, the specific steps are as follows: S21, performing monitoring image acquisition on a target accident space area to obtain accident image video stream data, and performing radar point cloud acquisition on the target accident space area to obtain accident area point cloud data; S22, acquiring an initial accident main body, and intercepting an accident area three-dimensional interaction model of which the initial accident main body is in a moving state in an interaction period to obtain a staged three-dimensional interaction model; Step S23, intercepting the staged three-dimensional interactive model according to frames to obtain a plurality of static staged three-dimensional models, and setting any two static staged three-dimensional models with adjacent intercepting time as a staged model comparison group to obtain a plurality of staged model comparison groups; s24, selecting a characteristic road main body in a stage model comparison group, and analyzing accident damage degree of an initial accident main body and the characteristic road main body in the stage model comparison group to acquire the invasion damage degree of the initial accident main body to the characteristic road main body; In the step S24, the specific steps are as follows: selecting a sample model comparison group, and naming two static staged three-dimensional models in the sample model comparison group as a first static three-dimensional model and a second static three-dimensional model according to the time sequence of interception; Randomly selecting a characteristic road body from the road bodies of which the second static three-dimensional model is contacted with the initial accident body; In a first static three-dimensional model, setting a space pixel point occupied by an initial accident main body as an initial space pixel point, setting a space pixel point occupied by a characteristic road main body as a characteristic space pixel point, acquiring the characteristic space pixel point which is in an overlapping state with the initial space pixel point to obtain a characteristic overlapping pixel point, and acquiring a model space region occupied by the characteristic overlapping pixel point to obtain a first main body invasion region; In the second static three-dimensional model, acquiring a model space region occupied by the feature overlapping pixel points to obtain a second main body intrusion region; Acquiring a new area of the second main body invasion area relative to the first main body invasion area, and obtaining the new invasion area of the characteristic road main body in the sample model comparison group; Acquiring newly-increased intrusion areas of the characteristic road body in each stage model comparison group, acquiring the volume of each newly-increased intrusion area to obtain a plurality of intrusion area volumes, and summing to obtain an area accumulated intrusion volume; collecting the moving speed of an initial accident main body in each sample model comparison group to obtain a plurality of moving speeds of the invading bodies; setting the volume of an intrusion region corresponding to the same newly-added intrusion region as A1, setting the cumulative intrusion volume of the region as A2, setting the moving speed of an intrusion body as A3, calculating A3 x (A1/A2) to obtain the weighted intrusion speed of the newly-added region, obtaining the weighted intrusion speed of the newly-added region corresponding to each newly-added intrusion region, summing the weighted intrusion speeds of the feature road main body by the initial accident main body, and obtaining the road speed limit corresponding to the target accident space region to obtain the regional road speed limit; Calculating the weighted invasion speed, regional road speed limit and regional accumulated invasion volume to obtain the invasion damage degree of the initial accident main body to the characteristic road main body Step S25, obtaining the invasion damage degree of each traffic accident main body to the characteristic road main body to obtain a plurality of invasion damage degrees; Step S26, respectively acquiring the regional accumulated intrusion volume of each traffic accident main body to the characteristic road main body, and summing to obtain the main body total intrusion volume; Step S27, setting the intrusion damage degree corresponding to the same traffic accident main body as B1, setting the regional accumulated intrusion volume as B2, setting the main body total intrusion volume as B3, calculating the weighted regional damage degree, and summing the obtained multiple weighted regional damage degrees to obtain the accident damage degree corresponding to the characteristic road main body; step S28, obtaining accident damage degree corresponding to each road main body to obtain accident damage evaluation data; s3, predicting the congestion time length of the target accident space region according to the accident damage evaluation data, and carrying out road network diversion early warning according to the prediction result; in the step S3, the method further includes the following steps: Step S31, accident damage evaluation data are obtained, and the accident damage degree corresponding to each road main body is obtained according to the accident damage evaluation data, so that a plurality of target accident damage degrees are obtained; Step S32, acquiring historical traffic accidents occurring in the target accident space area, and screening the historical traffic accidents into first-type accidents to be matched and second-type accidents to be matched; Step S33, periodically matching the first type of accidents to be matched with the target traffic accidents by road network speed, and obtaining a plurality of matched consistent accidents according to the matching result; Step S34, obtaining historical congestion time lengths corresponding to each matched consistent accident, carrying out mean value calculation on the obtained historical congestion time lengths to obtain a first time length prediction characteristic value, carrying out standard deviation calculation on the obtained historical congestion time lengths to obtain a second time length prediction characteristic value, and calculating the sum of the first time length prediction characteristic value and the second time length prediction characteristic value to obtain a predicted congestion time length corresponding to a target traffic accident; and step 35, setting a reasonable congestion duration interval, if the predicted congestion duration is in the reasonable congestion duration interval, issuing a detour early warning to the road network analysis area, and if the predicted congestion duration is not in the reasonable congestion duration interval, issuing a detour early warning to the road network analysis area.
2. 2. The traffic accident road network event cross-modal correlation analysis method based on the large model according to claim 1, wherein in the step S1, the specific steps are as follows: Step S11, acquiring traffic accidents occurring at the current moment to obtain target traffic accidents, and setting a space area corresponding to the target traffic accidents as a target accident space area; step S12, acquiring an initial accident main body, analyzing the moving speed, and acquiring an accident starting time point according to an analysis result; step S13, acquiring a time point when each traffic accident main body is in a static state, obtaining a main body movement stop time point, sequencing the acquired main body movement stop time points according to time sequence, and setting the main body movement stop time point arranged at the tail position as an accident end time point; and S14, setting the time interval between the accident starting time point and the accident ending time point as a target accident time period.
3. 3. The traffic accident road network event cross-modal correlation analysis method based on the large model according to claim 2, wherein in the step S12, the specific steps are as follows: acquiring an accident main body which participates in the accident collision first in the target traffic accident to obtain an initial accident main body; Setting a time point when an initial accident subject enters a target accident space area as a first accident characteristic time point, setting a moving stop time point of the initial accident subject in the target accident space area as a second accident characteristic time point, and setting a time period between the first accident characteristic time point and the second accident characteristic time point as a track monitoring time period; creating a speed time coordinate system, collecting the real-time speed of an initial accident main body in a track monitoring period, and drawing a first time speed curve; acquiring a traffic accident main body which collides with the initial accident main body first to obtain the initial traffic accident main body, acquiring the real-time moving speed of the initial traffic accident main body in a track monitoring period, and drawing a second time speed curve; In the speed time coordinate system, a coordinate range covered by the track monitoring time period on the coordinate x-axis is set as a time period coordinate range, a speed monitoring window is set, and a coordinate position of the speed monitoring window in the time period coordinate range is set as a characteristic coordinate position.
4. 4. The traffic accident road network event cross-modal correlation analysis method based on the large model according to claim 3, wherein in the step S12, the specific steps are as follows: Acquiring the ordinate of the intersection point of the left edge of the speed monitoring window and the first time speed curve to obtain a first characteristic speed value, acquiring the ordinate of the intersection point of the right edge of the speed monitoring window and the first time speed curve to obtain a second characteristic speed value, calculating the difference value between the first characteristic speed value and the second characteristic speed value, and taking an absolute value of the obtained difference value to obtain a first curve window speed difference; Acquiring the ordinate of the intersection point of the left edge of the speed monitoring window and the second time speed curve to obtain a third characteristic speed value, acquiring the ordinate of the intersection point of the right edge of the speed monitoring window and the second time speed curve to obtain a fourth characteristic speed value, calculating the difference value between the third characteristic speed value and the fourth characteristic speed value, and taking an absolute value of the obtained difference value to obtain a speed difference of the second curve window; calculating the sum of the speed difference of the first curve window and the speed difference of the second curve window to obtain the speed difference of the accident main body window corresponding to the speed monitoring window; Sliding and traversing the time-domain coordinate range by using the speed monitoring window, acquiring the speed differences of the accident main body windows corresponding to the speed monitoring windows of different coordinate positions according to the traversing result, comparing the values of the speed differences of the obtained accident main body windows, marking the speed monitoring window corresponding to the speed difference of the maximum accident main body window as a collision time window, and acquiring the intermediate value of the time period covered by the collision time window to obtain the accident starting time point.
5. 5. The traffic accident road network event cross-modal correlation analysis method based on the large model according to claim 1, wherein in the step S32, the method further comprises the steps of: a sample historical accident is selected at will from the obtained historical traffic accidents, the accident damage degree corresponding to each road main body of the sample historical accident is obtained, and the accident damage degree is named as the historical accident damage degree; Aiming at the same road main body, setting the damage degree of a target accident as C1, setting the damage degree of a historical accident as C2, calculating |C1-C2|/C1 to obtain damage deviation ratios corresponding to the road main bodies, obtaining the damage deviation ratio corresponding to each road main body, and summing to obtain the accident damage deviation degree of a sample historical accident and a target traffic accident; acquiring accident damage deviation degree of each historical traffic accident and a target traffic accident, and setting an accident damage deviation degree reference interval; If the accident damage deviation degree is in the accident damage deviation degree reference interval, the corresponding historical traffic accidents are classified into first type accidents to be matched, and if the accident damage deviation degree is not in the accident damage deviation degree reference interval, the corresponding historical traffic accidents are classified into second type accidents to be matched.
6. 6. The traffic accident road network event cross-modal correlation analysis method based on the large model according to claim 1, wherein in the step S33, the method further comprises the steps of: setting a road network analysis area at the periphery of a target accident space area, and acquiring each passing road in the road network analysis area to acquire the average moving speed of the vehicles of each passing road in a target accident time period to acquire a plurality of target vehicle moving speeds; selecting a sample accident to be matched from the first type accidents to be matched, and acquiring the average moving speed of the vehicles of each passing road in the accident time period corresponding to the sample accident to be matched to acquire a plurality of sample vehicle moving speeds; aiming at the same traffic road, setting the moving speed of a target vehicle as D1, setting the moving speed of a sample vehicle as D2, calculating the absolute value D1-D2 absolute value/D1 to obtain the speed deviation ratio corresponding to the traffic road, respectively obtaining the speed deviation ratio corresponding to each traffic road, and summing to obtain the road network speed deviation degree of the accident to be matched with the sample and the target traffic accident; obtaining road network vehicle speed deviation degree of each first type of accident to be matched and a target traffic accident, and setting a vehicle speed deviation degree reference interval; If the road network vehicle speed deviation is in the vehicle speed deviation reference interval, the corresponding first type accidents to be matched are classified as matching consistent accidents, and if the road network vehicle speed deviation is not in the vehicle speed deviation reference interval, the corresponding first type accidents to be matched are classified as non-matching consistent accidents.

Description

Traffic accident road network event cross-modal correlation analysis method based on large model Technical Field The invention belongs to the field of public transportation, relates to a large model technology, and in particular relates to a traffic accident road network event cross-modal correlation analysis method based on a large model. Background 1. When the existing road network event cross-mode association analysis method is used for carrying out traffic accident destructive analysis, accident damage inventory is carried out on an accident scene in a manual inspection mode, accident period collection is not carried out on the basis of the change of the relative movement speed of an accident participation main body, an accident area three-dimensional interaction model is not created on an accident area in an accident period, the intrusion volume and the intrusion speed of the accident main body to the road main body cannot be collected in real time according to the accident area three-dimensional interaction model, so that accident damage evaluation cannot be carried out on a target accident space area according to the accident damage evaluation, and hysteresis exists in a traffic accident evaluation process; 2. The existing road network event cross-modal correlation analysis method generally carries out bypass early warning on a regional road network based on the actual processing progress of traffic accidents, accident damage degree corresponding to a target traffic accident and road network speed are not matched with historical traffic accidents of the region in a consistent manner, and the target traffic accident cannot be predicted for congestion duration based on the actual congestion duration of the matched historical accidents, so that the bypass early warning lacks timeliness and accuracy. Therefore, we propose a traffic accident road network event cross-modal correlation analysis method based on a large model. Disclosure of Invention Aiming at the defects of the prior art, the invention aims to provide a traffic accident road network event cross-modal correlation analysis method based on a large model, and aims to improve the accuracy and timeliness of the traffic accident road network event cross-modal correlation analysis method. In order to achieve the purpose, the invention adopts the following technical scheme that the traffic accident road network event cross-modal association analysis method based on the large model comprises the following steps: Step S1, acquiring a target traffic accident, analyzing the moving speed of a participated subject corresponding to the target traffic accident, and obtaining a target accident time period according to an analysis result; Step S2, carrying out three-dimensional modeling on a target accident space region in a target accident time period to obtain an accident region three-dimensional interaction model, carrying out accident damage evaluation on the target accident space region according to the accident region three-dimensional interaction model, and obtaining accident damage evaluation data according to an evaluation result; And step S3, predicting the congestion time length of the target accident space region according to the accident damage evaluation data, and carrying out road network diversion early warning according to the prediction result. Further, in the step S1, the specific steps are as follows: Step S11, acquiring traffic accidents occurring at the current moment to obtain target traffic accidents, and setting a space area corresponding to the target traffic accidents as a target accident space area; Step S12, analyzing the moving speed of the participated subject corresponding to the target traffic accident, and acquiring an accident starting time point according to an analysis result; Step S13, respectively acquiring accident subjects participating in a target traffic accident to obtain a plurality of traffic accident subjects, respectively acquiring time points of each traffic accident subject in a static state to obtain subject movement stop time points, sequencing the acquired plurality of subject movement stop time points according to time, and setting the last subject movement stop time point as an accident end time point; and S14, setting the time interval between the accident starting time point and the accident ending time point as a target accident time period. Further, in the step S12, the specific steps are as follows: acquiring an accident main body which participates in the accident collision first in the target traffic accident to obtain an initial accident main body; Setting a time point when an initial accident subject enters a target accident space area as a first accident characteristic time point, setting a moving stop time point of the initial accident subject in the target accident space area as a second accident characteristic time point, and setting a time period between the first accident characteristic time