Traffic accident warning and triangle warning sign deployment system based on AI

Abstract

The present invention relates to the field of smart transportation technology. The present invention discloses an AI-based traffic accident warning and triangle warning sign deployment system, including calculating a static influence value based on static feature data, calculating a historical influence value based on historical feature data, determining an initial node sequence according to the static influence value and the historical influence value, inputting an event graph and a G-layer feature graph into a pre-built dynamic prediction model, obtaining a dynamic adjustment factor, and finally adjusting the initial node sequence according to the dynamic adjustment factor to obtain a target node sequence. The present invention adopts diversified data and distinguishes static information from dynamic information. First, an initial node sequence is generated according to the static feature data and the historical feature data, and then a dynamic adjustment factor is determined through the dynamic feature data. Finally, the initial node sequence is adjusted according to the dynamic adjustment factor to obtain a target node sequence, thereby further improving the accuracy of the prediction result.

Description

Technical Field

[0001] The present invention relates to the field of intelligent transportation technology, and more specifically, to an AI-based traffic accident warning and triangle warning sign deployment system. Background Art

[0002] As people's living standards continue to improve, the number of cars in my country has increased year by year, and traffic accidents and vehicle breakdowns occur from time to time. Generally speaking, warning signs need to be placed on roads where accidents often occur to remind drivers to pay attention to road safety. However, since the dynamic characteristics of roads (such as traffic volume) are constantly changing, artificial intelligence technology is needed to provide traffic accident warnings and reasonably select and deploy the locations of traffic warning signs. Road traffic safety assessment has gradually introduced big data, smart cities, data mining technologies, etc. The application of these new technologies can conduct targeted analysis of road traffic ecology. In the existing technology, although traffic accidents have been predicted through technologies such as big data, there are still certain problems.

[0003] For example, a Chinese patent with publication number CN118230556A provides a traffic parameter prediction method, device and equipment, which constructs a space-time graph model of the road network structure based on static information and real-time dynamic information, extracts spatial features from the space-time graph model to obtain the real-time spatial features of the road network structure, extracts time features from the space-time graph model to obtain the real-time time features of the road network structure, predicts traffic parameters of the road network structure based on the real-time spatial features and real-time time features, and obtains the traffic parameters of the road network structure at a preset future moment.

[0004] Although the above patent also predicts traffic parameters, the data used is not diversified enough. In addition, since static information does not have strong real-time characteristics, while dynamic information has strong real-time characteristics, and the difference between dynamic information and static information is large, the above patent predicts static information and dynamic information about traffic together, which undoubtedly causes inaccurate prediction results.

[0005] In view of this, the present invention proposes an AI-based traffic accident warning and triangle warning sign deployment system to solve the above problems. Summary of the invention

[0006] In order to overcome the above-mentioned defects of the prior art, the present invention provides an AI-based traffic accident warning and triangle warning sign deployment system.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] AI-based traffic accident warning and triangle warning sign deployment system, including:

[0009] Data acquisition module: used to acquire the road network data of the area to be monitored, determine M nodes in the area to be monitored based on the road network data, and respectively acquire static feature data, dynamic feature data and historical feature data corresponding to the M nodes, wherein the dynamic feature data at least includes descriptive text data and real-time image data, and M is an integer greater than 0;

[0010] The first processing module is used to calculate the static influence value based on the static feature data, calculate the historical influence value based on the historical feature data, and determine the initial node sequence according to the static influence value and the historical influence value;

[0011] The second processing module is used to construct an event graph based on the descriptive text data, obtain a G-layer feature graph based on the real-time image data, input the event graph and the G-layer feature graph into the pre-built dynamic prediction model, and obtain a dynamic adjustment factor, where G is an integer greater than 1:

[0012] Adjustment module: used to adjust the initial node sequence according to the dynamic adjustment factor to obtain the target node sequence, and determine the corresponding warning sign position according to the target node sequence.

[0013] Furthermore, the method for obtaining the G-layer feature map based on real-time image data includes:

[0014] According to the preset G extraction depths and feature pyramid network, the real-time image data is processed for feature extraction to obtain the G-layer feature map corresponding to the real-time image data. Figure 1 One to one correspondence.

[0015] Furthermore, the method for constructing an event graph based on the description text data includes:

[0016] Keyword recognition is performed on the description text data to obtain H keywords, and the H keywords are taken as objects. The corresponding relationships between the H objects are determined, and the H objects are taken as nodes and the corresponding relationships between the H objects are taken as edges to construct an event graph.

[0017] Furthermore, the static feature data includes node distance, number of road intersections, standard speed of passing, and waiting time of passing. The method for calculating the static influence value based on the static feature data includes:

[0018]

[0019] Where SIV is the static impact value, WTP is the waiting time, NDE is the node distance, NRI is the number of road intersections, SPD is the standard speed, log2[·] is the logarithmic function with base 2, cosh(·) is the hyperbolic cosine function, sin -1 (·) is the inverse sine function, exp(·) is the exponential function with e as the base, and e and k are both constants greater than 0.

[0020] Furthermore, the historical characteristic data includes the number of accidents, the number of peak hours, the accident response time, and the duration of the time period. The method for calculating the historical impact value based on the historical characteristic data includes:

[0021]

[0022] Where HIV is the historical impact value, NAS is the number of accidents, NPS is the number of peak hours, DUP is the duration of the period, ART is the accident response time, and tan -1 (·) is the inverse tangent function, cot -1 (·) is the inverse cotangent function, cos -1 (·) is the arccosine function.

[0023] Furthermore, the method for determining the initial node sequence according to the static influence value and the historical influence value includes:

[0024] The sum of the static influence value and the historical influence value is calculated according to a preset ratio as the initial influence value, and the initial influence values ​​corresponding to the M nodes are arranged in descending order to obtain an initial node sequence.

[0025] Furthermore, the method for constructing the dynamic prediction model includes:

[0026] A sample data set is obtained, wherein the sample data set includes a historical event graph, a historical feature graph, and a historical dynamic adjustment factor. The sample data set is divided into a sample training set and a sample test set, and a regression network is constructed. The historical event graph and the historical feature graph in the sample training set are used as input data of the regression network, and the historical dynamic adjustment factor in the sample training set is used as output data of the regression network. The regression network is trained to obtain an initial regression network for predicting the real-time dynamic adjustment factor. The initial regression network is tested using the sample test set, and the initial regression network that satisfies a preset error value is output as a dynamic prediction model.

[0027] Furthermore, the method of adjusting the initial node sequence according to the dynamic adjustment factor to obtain the target node sequence includes:

[0028] The initial influence value of each node in the initial node sequence is multiplied by the dynamic adjustment factor to obtain the target influence value, and then the nodes are arranged in descending order according to the target influence values ​​to obtain the target node sequence.

[0029] The AI-based traffic accident warning and triangle warning sign deployment method is implemented based on the above-mentioned AI-based traffic accident warning and triangle warning sign deployment system, including:

[0030] Obtain road network data of the area to be monitored, determine M nodes in the area to be monitored based on the road network data, and respectively obtain static feature data, dynamic feature data, and historical feature data corresponding to the M nodes, wherein the dynamic feature data at least includes descriptive text data and real-time image data, and M is an integer greater than 0;

[0031] Calculate the static influence value based on the static feature data, calculate the historical influence value based on the historical feature data, and determine the initial node sequence according to the static influence value and the historical influence value;

[0032] An event graph is constructed based on the description text data, a G-layer feature graph is obtained based on the real-time image data, and the event graph and the G-layer feature graph are input into a pre-constructed dynamic prediction model to obtain a dynamic adjustment factor, where G is an integer greater than 1;

[0033] The initial node sequence is adjusted according to the dynamic adjustment factor to obtain the target node sequence, and the corresponding warning sign position is determined according to the target node sequence.

[0034] A computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed, implements the above-mentioned AI-based traffic accident warning and triangle warning sign deployment method.

[0035] Compared with the prior art, the present invention has the following beneficial effects:

[0036] The present invention first calculates the static influence value based on the static feature data, calculates the historical influence value based on the historical feature data, determines the initial node sequence according to the static influence value and the historical influence value, constructs the event graph based on the descriptive text data, obtains the G-layer feature graph based on the real-time image data, inputs the event graph and the G-layer feature graph into the pre-built dynamic prediction model, obtains the dynamic adjustment factor, and finally adjusts the initial node sequence according to the dynamic adjustment factor to obtain the target node sequence, and determines the corresponding warning sign position according to the target node sequence. The present invention adopts diversified data and distinguishes between static information and dynamic information, first generates the initial node sequence according to the static feature data and the historical feature data, and then determines the dynamic adjustment factor through the dynamic feature data, and finally adjusts the initial node sequence according to the dynamic adjustment factor to obtain the target node sequence, thereby further improving the accuracy of the prediction result. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is a structural schematic diagram of the traffic accident warning and triangle warning sign deployment system based on AI in the present invention;

[0038] Figure 2 This is a flow chart of the AI-based traffic accident warning and triangle warning sign deployment method of the present invention;

[0039] Figure 3 is a schematic diagram of an event graph in the present invention;

[0040] Figure 4 A schematic diagram of a computer-readable storage medium in the present invention.

[0041] Reference numerals:

[0042] 11. Node; 12. Edge. DETAILED DESCRIPTION

[0043] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0044] Example 1

[0045] See also Figure 1 As shown, this embodiment discloses an AI-based traffic accident warning and triangle warning sign deployment system, including:

[0046] Data acquisition module: used to acquire the road network data of the area to be monitored, determine M nodes in the area to be monitored based on the road network data, and respectively acquire static feature data, dynamic feature data and historical feature data corresponding to the M nodes, wherein the dynamic feature data at least includes descriptive text data and real-time image data, and M is an integer greater than 0;

[0047] In this embodiment, the road network data includes a detailed road network map and geographic information data obtained from a map system, and also includes remote sensing images taken from drones, etc. The nodes mentioned above include intersections and turning points of roads. The M nodes in the area to be monitored are determined based on the road network data. They can be directly obtained through analysis tools in existing map systems (such as OpenStreetMap and Baidu Maps). The above method is a prior art and this embodiment will not go into details.

[0048] It should be noted that the above-mentioned static feature data include but are not limited to node distance, number of road intersections, standard speed and waiting time. The node distance refers to the distance of the node from other nodes within a preset range, which can be the total distance to other nodes or the average distance to other nodes. For example, a circular area with a diameter of R and a target node as the center is used as the preset range. The nodes falling into the circular area need to calculate the distance from the target node, and the sum of the distances is used as the node distance corresponding to the target node; the standard speed refers to the average speed through the node. For example, taking the node as a turning point of the road, a corresponding sensor can be set to calculate the average speed through the turning point within a certain period of time as the standard speed.

[0049] In the above, the number of road intersections refers to the number of roads passing through the node. For example, the intersection formed by the intersection of three roads has a corresponding road intersection number of 3. When the node is a turning point of the road, the road intersection number is usually 1. The waiting time for passing refers to the maximum time restricted by the node. Taking the node as an intersection as an example, the corresponding waiting time for passing is the red light waiting time. It can be understood that if the node is a turning point of the road, the corresponding waiting time for passing can be zero.

[0050] In this embodiment, the dynamic feature data includes at least descriptive text data and real-time image data. The descriptive text data represents text data describing the real-time traffic status of the node, for example, "intersection B is in peak hours, traffic flow is heavy, and the length of the vehicle queue exceeds 200 meters." The real-time image data refers to the real-time image set obtained by the camera set at the node. The above-mentioned descriptive text data can be obtained in real time through the existing navigation system.

[0051] Specifically, historical characteristic data include but are not limited to the number of accidents, the number of peak hours, the accident response time, and the duration of the time period. The number of accidents can be the average number of accidents that occur at the node per month, or the number of accidents that occur in several months; similarly, the number of peak hours can be the number of peak hours that occur every day, or the number of peak hours that occur in several days; the accident response time represents the time required from the occurrence of the accident to the arrival of rescue. It can be understood that the longer the accident response time, the greater the degree of impact of the accident on the node, and the greater the probability of subsequent traffic accidents at the node; the duration of the time period refers to the duration of the peak hour. The above historical characteristic data are continuously recorded through logs and stored in the system.

[0052] The first processing module is used to calculate the static influence value based on the static feature data, calculate the historical influence value based on the historical feature data, and determine the initial node sequence according to the static influence value and the historical influence value;

[0053] The method for calculating the static influence value based on static feature data includes:

[0054]

[0055] Where SIV is the static impact value, WTP is the waiting time, NDE is the node distance, NRI is the number of road intersections, SPD is the standard speed, log2[·] is the logarithmic function with base 2, cosh(·) is the hyperbolic cosine function, sin -1 (·) is the inverse sine function, exp(·) is the exponential function with e as the base, and e and k are both constants greater than 0.

[0056] In this embodiment, the waiting time and the number of road intersections are used as an example. The longer the waiting time, the more serious the congestion of the node and the greater the probability of an accident. The more road intersections there are, the more complex the node and the greater the probability of an accident. It can be concluded from the above that the static impact value is positively correlated with the probability of an accident at the node.

[0057] Methods for calculating historical impact values ​​based on historical feature data include:

[0058]

[0059] Where HIV is the historical impact value, NAS is the number of accidents, NPS is the number of peak hours, DUP is the duration of the period, ART is the accident response time, and tan -1 (·) is the inverse tangent function, cot -1 (·) is the inverse cotangent function, cos -1 (·) is the arccosine function.

[0060] In this embodiment, the number of accident occurrences and the duration of a time period are used as examples. The more accidents occur, the greater the probability of an accident occurring at the corresponding node. Similarly, the longer the time period lasts, the more severe the congestion at the corresponding node is, and the greater the probability of an accident occurring. From the above content, it can be concluded that the historical impact value is positively correlated with the probability of an accident occurring at the node.

[0061] The method for determining the initial node sequence according to the static influence value and the historical influence value includes:

[0062] The sum of the static influence value and the historical influence value is calculated according to a preset ratio as the initial influence value, and the initial influence values ​​corresponding to the M nodes are arranged in descending order to obtain an initial node sequence.

[0063] It should be noted that the preset ratio refers to the weight between the static impact value and the historical impact value. The preset ratio can be determined by expert experience. The descending order in the above is only an example. The initial node sequence represents the arrangement of the probability of accidents occurring at the nodes.

[0064] The second processing module is used to construct an event graph based on the descriptive text data, obtain a G-layer feature graph based on the real-time image data, input the event graph and the G-layer feature graph into the pre-built dynamic prediction model, and obtain a dynamic adjustment factor, where G is an integer greater than 1:

[0065] Methods for constructing event graphs based on descriptive text data include:

[0066] Keyword recognition is performed on the description text data to obtain H keywords, and the H keywords are taken as objects. The corresponding relationships between the H objects are determined, and the H objects are taken as nodes and the corresponding relationships between the H objects are taken as edges to construct an event graph.

[0067] It is understandable that in the prior art, there are many technologies specifically used for keyword recognition. For example, Word2Vec can capture the semantic similarity of words and extract keywords through clustering or other methods. YAKE is an unsupervised keyword extraction method that focuses on single-document keyword extraction. It can also use existing pre-trained language models to perform transfer learning to generate the required keyword extraction model. This embodiment will not go into details.

[0068] It should be noted that the descriptive text data represents text data describing the real-time traffic status of the node, which will be specifically explained below in conjunction with the following content.

[0069] Taking the node as an intersection as an example, the description text includes three description sub-texts, which are:

[0070] The first description sub-text: "Before the rush hour begins, traffic at intersection B is smooth and vehicles can pass quickly."

[0071] The second description sub-text: "Intersection B is at the beginning of the rush hour, traffic volume increases, and vehicles begin to queue up."

[0072] The third description sub-text: "During the middle and late peak hours, traffic congestion is severe at intersection B, with the length of the vehicle queue exceeding 200 meters, and it takes a long time to wait through the intersection."

[0073] Continuing from the above, the description text includes three description sub-texts for exemplary explanation, and keywords are extracted from the three description sub-texts respectively. The keywords extracted from the first description sub-text are "intersection B", "before peak hour" and "smooth", the keywords extracted from the second description sub-text are "intersection B", "early peak hour", "queue" and "waiting", and the keywords extracted from the third description sub-text are "intersection B", "mid-to-late peak hour", "congestion" and "long waiting". Then, there is a corresponding relationship between "before peak hour" and "smooth", a corresponding relationship between "early peak hour" and "queue" and "waiting", and a corresponding relationship between "mid-to-late peak hour" and "congestion" and "long waiting". It can be understood that the common keyword "intersection B" in the above three description sub-texts is used as the starting point to construct an event graph, such as Figure 3 As shown, there are nodes 11 and edges 12. It can be understood that the node 11 here represents the keywords in the above content, and the edge 12 represents the relationship between the keywords.

[0074] The method for obtaining the G-layer feature map based on real-time image data includes:

[0075] According to the preset G extraction depths and feature pyramid network, the real-time image data is processed for feature extraction to obtain the G-layer feature map corresponding to the real-time image data. Figure 1 One to one correspondence.

[0076] It should be noted that the Feature Pyramid Network is a deep learning model structure for computer vision tasks. Its main feature is to enhance the model's detection ability for objects of different scales by constructing a multi-scale feature map pyramid from high resolution to low resolution. The Feature Pyramid Network uses a multi-layer feature map of a basic convolutional neural network (such as ResNet) to extract features from shallow to deep layers. These feature maps have different resolutions and can capture information of different scales.

[0077] Taking G as 3 and the real-time image data as an intersection image as an example, three layers of feature maps will be generated. The first layer of feature maps captures low-level edge and texture information, including low-level detail information such as lane signs, pedestrians, and vehicle contours. The second layer of feature maps captures mid-level shapes and partial object information, including relatively complete mid-level structural information such as vehicles, pedestrians, and traffic lights. The third layer of feature maps captures high-level semantic information, including high-level semantic information such as the layout of the entire intersection and traffic flow patterns.

[0078] The construction method of dynamic prediction model includes:

[0079] A sample data set is obtained, wherein the sample data set includes a historical event graph, a historical feature graph, and a historical dynamic adjustment factor. The sample data set is divided into a sample training set and a sample test set, and a regression network is constructed. The historical event graph and the historical feature graph in the sample training set are used as input data of the regression network, and the historical dynamic adjustment factor in the sample training set is used as output data of the regression network. The regression network is trained to obtain an initial regression network for predicting the real-time dynamic adjustment factor. The initial regression network is tested using the sample test set, and the initial regression network that satisfies a preset error value is output as a dynamic prediction model. The initial regression network is preferably a deep neural network model.

[0080] It can be understood that the historical event graph mentioned above is constructed from historical description text data. Similarly, the historical feature graph mentioned above is extracted from historical image data, and the historical dynamic adjustment factor is based on expert knowledge to grade the congestion level of the node according to actual conditions, so as to set the corresponding dynamic adjustment factor according to the congestion level. In this embodiment, the more serious the congestion level of the node, the larger the value of the corresponding dynamic adjustment factor.

[0081] Adjustment module: used to adjust the initial node sequence according to the dynamic adjustment factor to obtain the target node sequence, and determine the corresponding warning sign position according to the target node sequence.

[0082] The method of adjusting the initial node sequence according to the dynamic adjustment factor to obtain the target node sequence includes:

[0083] The initial influence value of each node in the initial node sequence is multiplied by the dynamic adjustment factor to obtain the target influence value, and then the nodes are arranged in descending order according to the target influence values ​​to obtain the target node sequence.

[0084] Specifically, each of the above nodes has a corresponding initial influence value and a dynamic adjustment factor, so the target influence value is obtained, and then arranged in descending order to obtain a target node sequence.

[0085] In this embodiment, a corresponding warning sign is set in the front section of the road of each node. The warning sign can be automatically raised or lowered or illuminated to remind the driver that the node ahead is congested and needs to drive carefully. In this embodiment, the larger the target impact value, the greater the probability of an accident occurring at the node. Therefore, a corresponding impact threshold can be set. When the target impact value corresponding to the node is greater than the impact threshold, the corresponding warning sign is raised or powered on to make the warning sign more eye-catching.

[0086] In this embodiment, M nodes in the monitored area are determined based on the road network data, and the static feature data, dynamic feature data and historical feature data corresponding to the M nodes are respectively obtained, the static influence value is calculated based on the static feature data, and the historical influence value is calculated based on the historical feature data. The initial node sequence is determined according to the static influence value and the historical influence value, an event graph is constructed based on the descriptive text data, a G-layer feature graph is obtained based on the real-time image data, the event graph and the G-layer feature graph are input into a pre-constructed dynamic prediction model, and a dynamic adjustment factor is obtained. Finally, the initial node sequence is adjusted according to the dynamic adjustment factor to obtain a target node sequence, and the corresponding warning sign position is determined according to the target node sequence. This embodiment adopts diversified data, and distinguishes static information from dynamic information. First, an initial node sequence is generated according to the static feature data and the historical feature data, and then the dynamic adjustment factor is determined by the dynamic feature data. Finally, the initial node sequence is adjusted according to the dynamic adjustment factor to obtain a target node sequence, which further improves the accuracy of the prediction result.

[0087] Example 2

[0088] like Figure 2 As shown, this embodiment provides an AI-based traffic accident warning and triangle warning sign deployment method based on Embodiment 1, including:

[0089] S10: Obtain road network data of the area to be monitored, determine M nodes in the area to be monitored based on the road network data, and respectively obtain static feature data, dynamic feature data, and historical feature data corresponding to the M nodes, wherein the dynamic feature data at least includes descriptive text data and real-time image data, and M is an integer greater than 0;

[0090] It should be noted that the above-mentioned static feature data include but are not limited to node distance, number of road intersections, standard speed and waiting time. The node distance refers to the distance of the node from other nodes within a preset range, which can be the total distance to other nodes or the average distance to other nodes. For example, a circular area with a diameter of R and a target node as the center is used as the preset range. The nodes falling into the circular area need to calculate the distance from the target node, and the sum of the distances is used as the node distance corresponding to the target node; the standard speed refers to the average speed through the node. For example, taking the node as a turning point of the road, a corresponding sensor can be set to calculate the average speed through the turning point within a certain period of time as the standard speed.

[0091] In the above, the number of road intersections refers to the number of roads passing through the node. For example, the intersection formed by the intersection of three roads has a corresponding road intersection number of 3. When the node is a turning point of the road, the road intersection number is usually 1. The waiting time for passing refers to the maximum time restricted by the node. Taking the node as an intersection as an example, the corresponding waiting time for passing is the red light waiting time. It can be understood that if the node is a turning point of the road, the corresponding waiting time for passing can be zero.

[0092] In this embodiment, the dynamic feature data includes at least descriptive text data and real-time image data. The descriptive text data represents text data describing the real-time traffic status of the node, for example, "intersection B is in peak hours, traffic flow is heavy, and the length of the vehicle queue exceeds 200 meters." The real-time image data refers to the real-time image set obtained by the camera set at the node. The above-mentioned descriptive text data can be obtained in real time through the existing navigation system.

[0093] Specifically, historical characteristic data include but are not limited to the number of accidents, the number of peak hours, the accident response time, and the duration of the time period. The number of accidents can be the average number of accidents that occur at the node per month, or the number of accidents that occur in several months; similarly, the number of peak hours can be the number of peak hours that occur every day, or the number of peak hours that occur in several days; the accident response time represents the time required from the occurrence of the accident to the arrival of rescue. It can be understood that the longer the accident response time, the greater the degree of impact of the accident on the node, and the greater the probability of subsequent traffic accidents at the node. The duration of the time period refers to the duration of the peak hour. The above historical characteristic data are continuously recorded through logs and stored in the system.

[0094] S20: Calculate a static influence value based on the static feature data, calculate a historical influence value based on the historical feature data, and determine an initial node sequence according to the static influence value and the historical influence value;

[0095] The method for calculating the static influence value based on static feature data includes:

[0096]

[0097] Where SIV is the static impact value, WTP is the waiting time, NDE is the node distance, NRI is the number of road intersections, SPD is the standard speed, log2[·] is the logarithmic function with base 2, cosh(·) is the hyperbolic cosine function, sin -1 (·) is the inverse sine function, exp(·) is the exponential function with e as the base, and e and k are both constants greater than 0.

[0098] Methods for calculating historical impact values ​​based on historical feature data include:

[0099]

[0100] Where HIV is the historical impact value, NAS is the number of accidents, NPS is the number of peak hours, DUP is the duration of the period, ART is the accident response time, and tan -1 (·) is the inverse tangent function, cot -1 (·) is the inverse cotangent function, cos -1 (·) is the arccosine function.

[0101] The method for determining the initial node sequence according to the static influence value and the historical influence value includes:

[0102] The sum of the static influence value and the historical influence value is calculated according to a preset ratio as the initial influence value, and the initial influence values ​​corresponding to the M nodes are arranged in descending order to obtain an initial node sequence.

[0103] It should be noted that the preset ratio refers to the weight between the static impact value and the historical impact value. The preset ratio can be determined by expert experience. The descending order in the above is only an example. The initial node sequence represents the arrangement of the probability of accidents occurring at the nodes.

[0104] S30: constructing an event graph based on the description text data, obtaining a G-layer feature graph based on the real-time image data, inputting the event graph and the G-layer feature graph into a pre-constructed dynamic prediction model, and obtaining a dynamic adjustment factor, where G is an integer greater than 1;

[0105] Methods for constructing event graphs based on descriptive text data include:

[0106] Keyword recognition is performed on the description text data to obtain H keywords, and the H keywords are taken as objects. The corresponding relationships between the H objects are determined, and the H objects are taken as nodes and the corresponding relationships between the H objects are taken as edges to construct an event graph.

[0107] It is understandable that in the prior art, there are many technologies specifically used for keyword recognition. For example, Word2Vec can capture the semantic similarity of words and extract keywords through clustering or other methods. YAKE is an unsupervised keyword extraction method that focuses on single-document keyword extraction. It can also use existing pre-trained language models to perform transfer learning to generate the required keyword extraction model. This embodiment will not go into details.

[0108] The method for obtaining the G-layer feature map based on real-time image data includes:

[0109] According to the preset G extraction depths and feature pyramid network, the real-time image data is processed for feature extraction to obtain the G-layer feature map corresponding to the real-time image data. Figure 1 One to one correspondence.

[0110] The construction method of dynamic prediction model includes:

[0111] A sample data set is obtained, wherein the sample data set includes a historical event graph, a historical feature graph, and a historical dynamic adjustment factor. The sample data set is divided into a sample training set and a sample test set, and a regression network is constructed. The historical event graph and the historical feature graph in the sample training set are used as input data of the regression network, and the historical dynamic adjustment factor in the sample training set is used as output data of the regression network. The regression network is trained to obtain an initial regression network for predicting the real-time dynamic adjustment factor. The initial regression network is tested using the sample test set, and the initial regression network that satisfies a preset error value is output as a dynamic prediction model. The initial regression network is preferably a deep neural network model.

[0112] S40: adjusting the initial node sequence according to the dynamic adjustment factor to obtain a target node sequence, and determining corresponding warning sign positions according to the target node sequence.

[0113] The method of adjusting the initial node sequence according to the dynamic adjustment factor to obtain the target node sequence includes:

[0114] The initial influence value of each node in the initial node sequence is multiplied by the dynamic adjustment factor to obtain the target influence value, and then the nodes are arranged in descending order according to the target influence values ​​to obtain the target node sequence.

[0115] Specifically, each of the above nodes has a corresponding initial influence value and a dynamic adjustment factor, so the target influence value is obtained, and then arranged in descending order to obtain a target node sequence.

[0116] In this embodiment, a corresponding warning sign is set in the front section of the road of each node. The warning sign can be automatically raised or lowered or illuminated to remind the driver that the node ahead is congested and needs to drive carefully. In this embodiment, the larger the target impact value, the greater the probability of an accident occurring at the node. Therefore, a corresponding impact threshold can be set. When the target impact value corresponding to the node is greater than the impact threshold, the corresponding warning sign is raised or powered on to make the warning sign more eye-catching.

[0117] Example 3

[0118] The present embodiment discloses an electronic device, including a power supply, an interface, a keyboard, a memory, a central processing unit, and a computer program stored in the memory and executable on the central processing unit. When the central processing unit executes the computer program, the AI-based traffic accident warning and triangle warning sign deployment method provided by the above methods is implemented. The interface includes a network interface and a data interface. The network interface includes a wired or wireless interface, and the data interface includes an input or output interface.

[0119] Since the electronic device introduced in this embodiment is an electronic device used to implement the AI-based traffic accident warning and triangle warning sign deployment method in the embodiment of this application, based on the AI-based traffic accident warning and triangle warning sign deployment method introduced in the embodiment of this application, the technical personnel of this field can understand the specific implementation of the electronic device of this embodiment and its various variations, so how the electronic device implements the method in the embodiment of this application will not be described in detail here. As long as the technical personnel of this field implement the electronic device used by the AI-based traffic accident warning and triangle warning sign deployment method in the embodiment of this application, it belongs to the scope of protection of this application.

[0120] Example 4

[0121] like Figure 4 As shown, the present embodiment discloses a computer-readable storage medium, on which a computer program is stored. When the computer program is executed, the above-mentioned AI-based traffic accident warning and triangle warning sign deployment method is implemented.

[0122] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters, weights and thresholds in the formula are set by technicians in this field according to actual conditions.

[0123] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented by software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website site, computer, server or data center to another website site, computer, server or data center through a wired network or a wireless network. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a tape), an optical medium (for example, a DVD) or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.

[0124] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in the present invention can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0125] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0126] In the several embodiments provided by the present invention, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of the units is only one, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0127] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0128] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0129] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art who is familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

[0130] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. The AI-based traffic accident warning and triangle warning sign deployment system is characterized by: include: Data acquisition module: used to acquire the road network data of the area to be monitored, determine M nodes in the area to be monitored based on the road network data, and respectively acquire static feature data, dynamic feature data and historical feature data corresponding to the M nodes, wherein the dynamic feature data at least includes descriptive text data and real-time image data, and M is an integer greater than 0; The first processing module is used to calculate the static impact value based on the static feature data, calculate the historical impact value based on the historical feature data, and determine the initial node sequence according to the static impact value and the historical impact value, wherein the static impact value is used to characterize the degree of influence of the static feature data on the node accident, and the historical impact value is used to characterize the degree of influence of the historical feature data on the node accident; The second processing module is used to construct an event graph based on the description text data, obtain a G-layer feature graph based on the real-time image data, input the event graph and the G-layer feature graph into a pre-built dynamic prediction model, and obtain a dynamic adjustment factor, where G is an integer greater than 1; Adjustment module: used to adjust the initial node sequence according to the dynamic adjustment factor to obtain the target node sequence, and determine the corresponding warning sign position according to the target node sequence; The method for obtaining a G-layer feature map based on real-time image data comprises: Perform feature extraction processing on the real-time image data according to the preset G extraction depths and feature pyramid network to obtain G-layer feature maps corresponding to the real-time image data, and the G extraction depths correspond to the G-layer feature maps one by one; The method for constructing an event graph based on description text data comprises: Perform keyword recognition on the description text data to obtain H keywords, take the H keywords as objects, determine the corresponding relationships between the H objects, take the H objects as nodes, and the corresponding relationships between the H objects as edges, so as to construct an event graph; The method for constructing the dynamic prediction model includes: Acquire a sample data set, wherein the sample data set includes a historical event graph, a historical feature graph, and a historical dynamic adjustment factor, divide the sample data set into a sample training set and a sample test set, construct a regression network, use the historical event graph and the historical feature graph in the sample training set as input data of the regression network, use the historical dynamic adjustment factor in the sample training set as output data of the regression network, train the regression network, obtain an initial regression network for predicting the real-time dynamic adjustment factor, test the initial regression network using the sample test set, and output the initial regression network that satisfies a preset error value as a dynamic prediction model; The method for determining the initial node sequence according to the static influence value and the historical influence value comprises: The sum of the static influence value and the historical influence value is calculated according to a preset ratio as the initial influence value, and the initial influence values ​​corresponding to the M nodes are arranged in descending order to obtain an initial node sequence; The method of adjusting the initial node sequence according to the dynamic adjustment factor to obtain the target node sequence includes: The initial influence value of each node in the initial node sequence is multiplied by the dynamic adjustment factor to obtain the target influence value, and then the nodes are arranged in descending order according to the target influence values ​​to obtain the target node sequence.

2. The AI-based traffic accident warning and triangle warning sign deployment system according to claim 1 is characterized in that: The static feature data include node distance, number of road intersections, standard speed and waiting time. The method for calculating the static influence value based on the static feature data includes: Where SIV is the static impact value, WTP is the waiting time, NDE is the node distance, NRI is the number of road intersections, SPD is the standard speed, log2[·] is the logarithmic function with base 2, cosh(·) is the hyperbolic cosine function, sin -1 (·) is the inverse sine function, exp(·) is the exponential function with e as the base, and e and k are both constants greater than 0.

3. The AI-based traffic accident warning and triangle warning sign deployment system according to claim 1 is characterized in that: The historical characteristic data includes the number of accidents, the number of peak hours, the accident response time, and the duration of the time period. The method for calculating the historical impact value based on the historical characteristic data includes: Where HIV is the historical impact value, NAS is the number of accidents, NPS is the number of peak hours, DUP is the duration of the period, ART is the accident response time, and tan -1 (·) is the inverse tangent function, cot -1 (·) is the inverse cotangent function, cos -1 (·) is the arccosine function.

4. An AI-based traffic accident warning and triangle warning sign deployment method, which is implemented based on the AI-based traffic accident warning and triangle warning sign deployment system according to any one of claims 1 to 3, and is characterized in that: include: Obtain road network data of the area to be monitored, determine M nodes in the area to be monitored based on the road network data, and respectively obtain static feature data, dynamic feature data, and historical feature data corresponding to the M nodes, wherein the dynamic feature data at least includes descriptive text data and real-time image data, and M is an integer greater than 0; Calculate the static influence value based on the static feature data, calculate the historical influence value based on the historical feature data, and determine the initial node sequence according to the static influence value and the historical influence value; An event graph is constructed based on the description text data, a G-layer feature graph is obtained based on the real-time image data, and the event graph and the G-layer feature graph are input into a pre-constructed dynamic prediction model to obtain a dynamic adjustment factor, where G is an integer greater than 1; The initial node sequence is adjusted according to the dynamic adjustment factor to obtain the target node sequence, and the corresponding warning sign position is determined according to the target node sequence.

5. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed, the AI-based traffic accident warning and triangle warning sign deployment method described in claim 4 is implemented.

Images