A macro-region vehicle early warning method and system based on safety risk field

Abstract

The present invention discloses a macro-region vehicle early warning method and system based on a safety risk field, including: S1, constructing a vehicle driving risk field model; S2, collecting vehicle surrounding environment information; S3, based on the vehicle surrounding environment information data collected in step S2, predicting the position coordinates and speed of the vehicle and surrounding vehicles in the future, and obtaining the position and speed data information of the vehicle and surrounding vehicles at each moment in the future time period; S4, based on the obtained position and speed data of the vehicle and surrounding vehicles, respectively calculating the driving risk field of the vehicle in each time period; S5, judging whether there will be interaction between vehicles according to the inverse of the time taken for the vehicle to collide normally, calculating the conflict risk field value and the comprehensive conflict risk field value; S6, real-time output of the comprehensive conflict risk field value obtained according to the calculation. The present invention can realize the risk level assessment of various situations that may occur on the road.

Description

Technical Field

[0001] The invention relates to a macro-region vehicle early warning method, belonging to the technical field of safe driving early warning. Background Art

[0002] While people enjoy the speed and convenience brought by cars, they also have to face the "side effect" of traffic accidents. Advanced car safety facilities are an indispensable guarantee for driving safety. Since the advent of cars, car manufacturers have been working hard to design cars with more complete safety facilities. However, if the driver has a weak sense of safe driving, does not understand or does not use these safety facilities correctly, or even violates the rules of driving safety, even if the safety configuration of the car is advanced, driving safety cannot be guaranteed. Therefore, automobile safe driving technology plays an important role in automobile safety. The problem that still exists in the implementation of the above device is that in order to maintain the safe driving of the vehicle, the existing vehicles often have various warning functions, such as anti-collision warning, bad weather, and road section fault warning in common map software, etc. CN 114212105 discloses a highly generalized interactive vehicle driving intention prediction method, which obtains the interactive scene information of the heading relationship information, reference route information and road structure information between the vehicle and the interactive vehicle, determines the interactive scene type, trains the interactive vehicle driving intention prediction model through pre-constructed training data, predicts the interactive vehicle driving intention through the trained interactive vehicle driving intention prediction model, and presents the prediction result through human-computer interaction. The existing technology can usually only warn of a certain dangerous situation in traffic, and lacks a unified indicator to evaluate the degree of danger of various situations that may occur on the road. Summary of the invention

[0003] The technical problem to be solved by the present invention is: a vehicle early warning method is needed to evaluate the danger level of various situations that may occur on the road.

[0004] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0005] A macro-region vehicle early warning method based on a safety risk field comprises the following steps:

[0006] S1. Construct a vehicle driving risk field model and use the risk field to represent the risk conditions that the vehicle is exposed to during operation;

[0007] S2, collecting vehicle surrounding environment information, including the distance between the vehicle and surrounding vehicles, the vehicle's geographic location, the size and outline of surrounding vehicles, and the vehicle's acceleration and direction;

[0008] S3, based on the vehicle surrounding environment information data collected in step S2, use the vehicle trajectory prediction model to predict the position coordinates and speed changes of the vehicle and surrounding vehicles in the future, and obtain the position and speed data information of the vehicle and surrounding vehicles at each moment in the future time period;

[0009] S4, based on the position and speed data of the vehicle and surrounding vehicles obtained in step S3, using the vehicle driving risk field model in step 1, respectively calculate the driving risk field of the vehicle in each time period;

[0010] S5. Determine whether there will be interaction between vehicles based on the inverse of the time taken for the vehicles to collide with each other during normal driving. If there is interaction between vehicles, the overlapping area of ​​multiple driving risk fields is the conflict risk field, and the conflict risk field value is calculated. The conflict risk field values ​​of multiple conflict risk fields in the area where the vehicles that have the interaction are located are added together to obtain the comprehensive conflict risk field value in the area at each time point in the changing time period.

[0011] S6. Outputting the comprehensive conflict risk field value calculated in step S5 in real time, the change of the comprehensive conflict risk field value is the change of the safety risk of vehicles in the area in the future period of time;

[0012] S7. According to the change in the safety risk of the vehicle in step S6, when the comprehensive conflict risk value of the area increases due to the change in the vehicle's motion state, an early warning is issued.

[0013] In the aforementioned macro-region vehicle early warning method based on safety risk field, in step S1, the vehicle driving risk field model formula is:

[0014]

[0015] In the formula, f i (x,y) is the risk field of vehicle i, M i is the equivalent mass of the vehicle, a is the acceleration of the vehicle, θ is the angle between vehicle i and the surrounding vehicles, p i (x, y, t) is the probability density of the vehicle at (x, y) at time t predicted by the trajectory prediction model, σ is the prediction time domain, Δt is the time interval between two predictions, γ(t) is the discount function, λ1 and λ2 are the undetermined coefficients 1 and 2 respectively;

[0016] M i =m i (1.566·10 -14 v 6.687 +0.3345)

[0017] In the formula, m i is the actual mass of vehicle i, v is the speed of vehicle i;

[0018]

[0019] In the formula, 0.5 means that the amount after each half-life is reduced to half of the original amount, t 1 / 2 is the half-life.

[0020] The aforementioned macro-region vehicle warning method based on the safety risk field, in step S2, collects the distance between the vehicle and surrounding vehicles, the vehicle's geographical location, the size and outline of surrounding vehicles, the acceleration and direction of the vehicle through millimeter wave radar, GPS, lidar, and combined inertial navigation equipment; the vehicles include the vehicle itself, the front and rear vehicles in the vehicle's lane, and the front and rear vehicles in the left and right lanes.

[0021] In the aforementioned macro-region vehicle early warning method based on the safety risk field, in step S3, the vehicle trajectory prediction model uses a GNN model, and the prediction process of the GNN model includes:

[0022] S31. Preprocess the input data, including:

[0023] S311 standardizes vehicle status information, wherein the vehicle status information includes position, speed, and acceleration;

[0024] S312 selects vehicles as nodes, defines edges according to the relative positions between vehicles, and constructs a graph, wherein the feature vector of each vehicle node includes the position, speed, acceleration, direction, and lane of the vehicle;

[0025] S32. Encoding the state information into a form that can be processed by the GNN model;

[0026] S33. Perform forward propagation through multiple GNN layers to update the representation of each vehicle node;

[0027] S34. Use the time module of the recurrent neural network to process the historical trajectory of the vehicle, capture the spatiotemporal characteristics of the vehicle, and the spatiotemporal characteristics include speed and position information to achieve vehicle spatiotemporal relationship modeling;

[0028] S35.GNN model combines the current state of the vehicle with the learned spatiotemporal features to predict the future position of the vehicle through the fully connected layer;

[0029] S36.GNN model outputs the predicted position of each vehicle in the future, including a series of coordinate points representing the predicted trajectory.

[0030] The aforementioned macro-region vehicle early warning method based on the safety risk field, in step S33, includes:

[0031] S331 Each vehicle node collects feature information of adjacent nodes, where the adjacent nodes represent surrounding vehicles;

[0032] S332 The current node updates its own state using the collected feature information and its own feature information;

[0033] S333 uses a multi-layer GNN structure to learn higher-order interactions between nodes and achieve recursive updates of the graph;

[0034] The structure of multi-layer GNN is:

[0035] First layer: The information of directly adjacent nodes is aggregated to the current node, and the current node obtains the information of the first-order adjacent nodes;

[0036] Second layer: The output of the adjacent nodes in the first layer is aggregated with the features of the adjacent nodes again, so that the current node obtains the information of the second-order nodes;

[0037] More layers: Each additional layer can integrate information from more distant nodes and capture higher-order interactions.

[0038] In the aforementioned macro-region vehicle early warning method based on the safety risk field, in step S34, the vehicle history trajectory processing process includes:

[0039] S341. Input historical trajectory data, including vehicle dynamic information and timestamp information;

[0040] S342. Convert the trajectory data into a series of time-ordered state vectors, where the state vectors are state information of the vehicle at consecutive time steps, including position and speed;

[0041] S343. Process the state vector in the embedding layer to reduce the dimension and extract abstract features;

[0042] S344. Input the embedded features into the RNN model, and the RNN model captures the dynamic changes in the time series data;

[0043] S345. Each time step of the RNN model outputs a new hidden state to predict the vehicle position at the next time step;

[0044] S346.RNN model generates an output sequence, which contains predictions of the future position of the vehicle.

[0045] The aforementioned macro-region vehicle early warning method based on the safety risk field, in step S35, includes the following steps:

[0046] S351. Fusing the spatial features obtained from the GNN layer with the temporal features obtained from the time module of the RNN model, wherein the fusion is achieved by a splicing operation;

[0047] S352. Through the fully connected layer of the GNN model, the fused features are converted into multiple prediction values ​​suitable for final prediction, and the prediction values ​​directly correspond to the specific representation of the future state of the vehicle, including the future position coordinates, speed and direction, and probability distribution;

[0048] S353. Use multiple layers of fully connected layers to build a deep network structure, which captures and integrates multi-level abstract features;

[0049] S354.The last fully connected layer is the output layer, which maps the final high-level features to predicted values, including the location coordinates of the vehicle in the next few time steps.

[0050] In the above-mentioned macro-region vehicle early warning method based on the safety risk field, in step S5, it is determined whether there will be interaction between vehicles according to the inverse of the time TTC taken by the vehicle from normal driving to collision. The interaction judgment formula is:

[0051]

[0052] In the formula, Interaction is the interaction between vehicles. is the inverse of the TTC of vehicle i, t0 is the prediction start time, and s is the time unit in seconds;

[0053] The calculation formula of conflict risk field value is:

[0054] CSF=∫∫f(x,y)g(x,y)...m(x,y)dxdy

[0055] In the formula, CSF is the conflict risk field value, f(x,y), g(x,y), and m(x,y) are the risk field values ​​of each vehicle that has an interactive impact;

[0056] The calculation formula for the comprehensive conflict risk field value is:

[0057] CSF total =∑CSF i

[0058] Where, CSF total is the comprehensive conflict risk field value in the region, CSF i is the i-th conflict risk field value.

[0059] A computer system comprises a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above method.

[0060] A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that the computer program / instruction implements the steps of the above method when executed by a processor.

[0061] The beneficial effects achieved by the present invention are as follows: the method of the present invention is based on the safety potential field theory, and makes early warnings for the safe driving of regional vehicles from a macroscopic perspective. The vehicle's surrounding environment information is collected through millimeter wave radar, GPS, laser radar, and combined inertial navigation equipment. Then, the position coordinates and speed changes of the vehicle and surrounding vehicles in the future are predicted, and the risk values ​​generated between vehicles are calculated. The risk values ​​of the areas where the vehicles that interact are located are added, and finally the safety of the vehicle during driving is obtained. Finally, when the vehicle's motion state changes (overtaking, lane changing, speed change) and causes the overall risk of the area to increase, the vehicle is given a real-time early warning, thereby solving the problem that the existing technology can usually only give early warnings for a certain dangerous situation in traffic, and lacks a unified indicator to evaluate the degree of danger of various situations that may occur on the road. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 This is a flow chart of a macro-region vehicle early warning method based on a safety risk field according to Example 1 of the present invention. DETAILED DESCRIPTION

[0063] 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.

[0064] Example 1

[0065] like Figure 1 As shown, this embodiment provides a macro-region vehicle early warning method based on a safety risk field, comprising the following steps:

[0066] S1. Construct a vehicle driving risk field model and use the risk field to represent the risk conditions that the vehicle is exposed to during operation;

[0067] S2, collecting vehicle surrounding environment information, including the distance between the vehicle and surrounding vehicles, the vehicle's geographic location, the size and outline of surrounding vehicles, and the vehicle's acceleration and direction;

[0068] S3, based on the vehicle surrounding environment information data collected in step S2, use the vehicle trajectory prediction model to predict the position coordinates and speed changes of the vehicle and surrounding vehicles in the future, and obtain the position and speed data information of the vehicle and surrounding vehicles at each moment in the future time period;

[0069] S4, based on the position and speed data of the vehicle and surrounding vehicles obtained in step S3, using the vehicle driving risk field model in step 1, respectively calculate the driving risk field of the vehicle in each time period;

[0070] S5. Determine whether there will be interaction between vehicles based on the inverse of the time taken for the vehicles to collide with each other in normal driving. If there is interaction between vehicles, the driving risk fields of the vehicles will overlap. The overlapping area of ​​multiple driving risk fields is the conflict risk field, and the conflict risk field value is calculated. The conflict risk field values ​​of multiple conflict risk fields in the area where the interactive vehicles are located are added together to obtain the comprehensive conflict risk field value in the area at each time in the changing time period.

[0071] S6. Outputting the comprehensive conflict risk field value calculated in step S5 in real time. The change of the comprehensive conflict risk field value is the change of the safety risk of vehicles in the area in the future. The larger the comprehensive conflict risk field value, the higher the risk level.

[0072] S7. According to the change in the safety risk of the vehicle in step S6, when the comprehensive conflict risk value of the area increases due to the change in the vehicle's motion state, an early warning is issued.

[0073] In step S1, a risk field is used to represent the risk situation that a vehicle is exposed to during operation. A risk field exists during the operation of a vehicle. When there is interaction between vehicles, the risk fields of multiple vehicles overlap to generate a conflict risk field, and vehicle warning information is generated based on the conflict risk field.

[0074] The vehicle driving risk field model formula is:

[0075]

[0076] In the formula, f i (x,y) is the risk field of vehicle i, M i is the equivalent mass of the vehicle, a is the acceleration of the vehicle, θ is the angle between vehicle i and the surrounding vehicles, p i (x, y, t) is the probability density of the vehicle at (x, y) at time t predicted by the trajectory prediction model, σ is the prediction time domain, Δt is the time interval between two predictions, γ(t) is the discount function, λ1 and λ2 are the undetermined coefficients 1 and 2 respectively;

[0077] M i =m i (1.566·10 -14 v 6.687 +0.3345)

[0078] In the formula, m iis the actual mass of vehicle i, v is the speed of vehicle i;

[0079]

[0080] In the formula, 0.5 means that the amount after each half-life is reduced to half of the original amount, t 1 / 2 is the half-life.

[0081] In step S2, the distance between the vehicle and surrounding vehicles, the geographic location of the vehicle, the size and outline of surrounding vehicles, the acceleration and direction of the vehicle are collected respectively through millimeter wave radar, GPS, laser radar, and combined inertial navigation equipment; the vehicles include the vehicle itself, the front and rear vehicles in the lane of the vehicle itself, and the front and rear vehicles in the left and right lanes.

[0082] In step S3, based on the vehicle surrounding environment information collected in step S2, a vehicle trajectory prediction model is used to predict the position coordinates and speed changes of the vehicle and surrounding vehicles in the future, so as to obtain the position and speed information of the vehicle at each moment in the future time period. The vehicle trajectory prediction model uses a GNN (Graph Neural Networks) model.

[0083] The prediction process of the GNN model includes inputting data, performing feature extraction and interactive learning, and finally generating future trajectory predictions, including the following steps:

[0084] S31. Preprocess the input data, including:

[0085] S311 standardizes the state information of the vehicle, wherein the state information of the vehicle includes position, speed, acceleration, etc., so that the model can better understand the above characteristics;

[0086] S312 selects vehicles as nodes, defines edges according to the relative positions between vehicles, and constructs a graph, wherein the feature vector of each vehicle node includes the position, speed, acceleration, direction, and lane of the vehicle;

[0087] S32. Encoding the state information into a form that can be processed by the GNN model;

[0088] S33. Perform forward propagation through multiple GNN layers to update the representation of each vehicle node, including:

[0089] S331 Each vehicle node collects feature information of adjacent nodes, where the adjacent nodes represent surrounding vehicles;

[0090] S332 The current node updates its own state using the collected feature information and its own feature information;

[0091] First, the state of the current node is updated. The GNN model collects information from the neighboring nodes of the current node. Then, the current node combines the aggregated neighboring node information with the current features or hidden states of its own node to update its own node state.

[0092] S333 uses a multi-layer GNN structure to learn higher-order interactions between nodes and achieve recursive updates of the graph;

[0093] In each layer of GNN, the features of the nodes are updated according to the information of the adjacent nodes, and the multi-layer structure allows each node to identify more distant nodes. The structure of the multi-layer GNN is:

[0094] First layer: The information of directly adjacent nodes is aggregated to the current node, and the current node obtains the information of the first-order adjacent nodes;

[0095] Second layer: The output of the adjacent nodes in the first layer is aggregated with the features of the adjacent nodes again, so that the current node obtains the information of the second-order nodes;

[0096] More layers: Each additional layer can integrate information from more distant nodes and capture higher-order interactions.

[0097] For example, suppose there is a simple road network where vehicle A is adjacent to vehicles B and C, and vehicle B is adjacent to vehicle D, but vehicles A and D are not directly adjacent. In a single-layer GNN, the updated state of vehicle A can only reflect the direct relationship with B and C, but in a two-layer GNN, the new state of vehicle A will also indirectly contain relationship information with vehicle D because the information has been captured in the update of vehicle B and passed to A.

[0098] S34. The time module of the recurrent neural network (RNN) is used to process the historical trajectory of the vehicle, capture the spatiotemporal characteristics of the vehicle, and realize the spatiotemporal relationship modeling of the vehicle. The vehicle historical trajectory processing process includes:

[0099] S341. Input historical trajectory data, including dynamic information such as vehicle speed and acceleration, and timestamp information;

[0100] S342. Convert the trajectory data into a series of time-ordered state vectors, where the state vectors are state information of the vehicle at consecutive time steps, including position, speed, etc.;

[0101] S343. Process the state vector in the embedding layer to reduce the dimension and extract abstract features;

[0102] S344. Input the embedded features into the RNN model, and the RNN model captures the dynamic changes in the time series data;

[0103] S345. Each time step of the RNN model outputs a new hidden state to predict the vehicle position at the next time step;

[0104] S346.RNN model generates an output sequence, wherein the output sequence includes a prediction of the future position of the vehicle;

[0105] The S35.GNN model combines the current state of the vehicle with the learned spatiotemporal features to predict the future position of the vehicle through a fully connected layer, including the following steps:

[0106] S351. Fusing the spatial features obtained from the GNN layer with the temporal features obtained from the time module of the RNN model, wherein the fusion is achieved by a splicing operation;

[0107] S352. Through the fully connected layer of the GNN model, the fused features are converted into multiple prediction values ​​suitable for final prediction. The prediction values ​​directly correspond to the specific representation of the future state of the vehicle, including the future position coordinates, speed and direction, and probability distribution, so as to adapt to the vehicle trajectory prediction, and the high-dimensional features can be mapped to the prediction space;

[0108] S353. Use multiple layers of fully connected layers to build a deep network structure. The fully connected layer captures and integrates multi-level abstract features to provide support for vehicle trajectory prediction. By introducing a deep network structure after feature fusion, the GNN model can not only learn the relationship between features, but also adjust the level and abstraction of feature representation, and finally generate accurate prediction results. The deep network structure design enables the GNN model to capture spatial relationships and understand time series characteristics. At the same time, the deep learning ability of the fully connected layer is used to integrate and transform this information to achieve the purpose of prediction;

[0109] S354. The last fully connected layer is the output layer, which maps the final high-level features to predicted values, including the location coordinates of the vehicle in the next few time steps;

[0110] S36.GNN model outputs the predicted position of each vehicle in the future, including a series of coordinate points representing the predicted trajectory.

[0111] In step S4, based on the position and speed data of the vehicle and surrounding vehicles obtained in step S3, the driving risk field model of the vehicle in step 1 is used to calculate the driving risk field of the vehicle in each time period respectively;

[0112] The model formula of the vehicle's driving risk field is:

[0113]

[0114] In the formula, f i(x,y) is the risk field of vehicle i, M i is the equivalent mass of the vehicle, a is the acceleration of the vehicle, θ is the angle between vehicle i and the surrounding vehicles, p i (x, y, t) is the probability density of the vehicle at (x, y) at time t predicted by the trajectory prediction model, σ is the prediction time domain, Δt is the time interval between two predictions, γ(t) is the discount function, λ1 and λ2 are the undetermined coefficient one and undetermined coefficient two, respectively.

[0115] M i =m i (1.566·10 -14 v 6.687 +0.3345)

[0116] Where M i is the equivalent mass of vehicle i, m i is the actual mass of vehicle i, and v is the speed of vehicle i.

[0117]

[0118] In the formula, γ(t) is the discount function, 0.5 means that the amount after each half-life is reduced to half of the original amount, and t 1 / 2 is the half-life.

[0119] In step S5, it is determined whether there will be interaction between vehicles based on the inverse of the time TTC (Time-To-Collision, TTC) taken for the vehicles to travel normally until a collision occurs. The interaction determination formula is:

[0120]

[0121] In the formula, Interaction is the interaction between vehicles. is the inverse of the TTC of vehicle i, t0 is the prediction start time, and s is the time unit in seconds;

[0122] If there is interaction between vehicles, the driving risk fields of the vehicles will overlap. The overlapping area of ​​multiple driving risk fields is the conflict risk field, and the conflict risk field value is calculated. The calculation formula of the conflict risk field value is:

[0123] CSF=∫∫f(x,y)g(x,y)...m(x,y)dxdy

[0124] In the formula, CSF is the conflict risk field value, f(x,y), g(x,y), and m(x,y) are the risk field values ​​of each vehicle that has an interactive impact;

[0125] The conflict safety field values ​​of multiple conflict risk fields in the area where the vehicles that interact are located are added together to obtain the comprehensive conflict safety field value (CSF) in the area at each time during the changing time period.

[0126] CSF total =∑CSF i

[0127] Where, CSF total is the comprehensive conflict risk field value in the region, CSF i is the i-th conflict risk field value.

[0128] The risk values ​​between interacting vehicles are calculated based on their own established coordinate system, and the position information of surrounding vehicles is converted into the coordinate system established based on the target vehicle's position.

[0129] In step S6, the comprehensive conflict risk field value calculated in step S5 is output in real time, and the change of the comprehensive conflict risk field value is the change of the safety risk of vehicles in the area in the future period of time.

[0130] In step S7, according to the change in the safety risk of the vehicle in step S6, when the comprehensive conflict risk value of the area increases due to the change in the vehicle's motion state, an early warning is given through sound prompts, vibration prompts, and display prompts to remind the driver to change the vehicle's motion state in advance to avoid the occurrence of dangerous events and continue driving in the original state. When the comprehensive risk of the area will not increase, the vehicle is informed that it can change lanes, etc.; the vehicle motion state includes overtaking, changing lanes, and changing speed.

[0131] The early warning method of this embodiment is based on the safety potential field theory, and makes early warnings for the safe driving of regional vehicles from a macro perspective. It can collect information about the vehicle's surrounding environment through millimeter wave radar, GPS, laser radar, and combined inertial navigation equipment, and then predict the position coordinates and speed changes of the vehicle and surrounding vehicles in the future, calculate the risk value generated between vehicles, add the risk values ​​of the areas where the vehicles that interact are located, and finally derive the safety of the vehicle during driving. Finally, when the vehicle's motion state changes (overtaking, lane changing, speed change) and causes the overall risk of the area to increase, the vehicle is warned in real time, thereby solving the problem that the existing technology can usually only warn of a certain dangerous situation in traffic, and lacks a unified indicator to assess the degree of danger of various situations that may occur on the road.

[0132] Example 2

[0133] A computer system comprises a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above method.

[0134] Example 3

[0135] A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that the computer program / instruction implements the steps of the above method when executed by a processor.

[0136] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0137] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0138] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0139] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

Claims

1. A macro-region vehicle early warning method based on a safety risk field, characterized in that: The following steps are involved: S1. Construct a vehicle driving risk field model and use the risk field to represent the risk conditions that the vehicle is exposed to during operation; S2, collecting vehicle surrounding environment information, including the distance between the vehicle and surrounding vehicles, the vehicle's geographic location, the size and outline of surrounding vehicles, and the vehicle's acceleration and direction; S3, based on the vehicle surrounding environment information data collected in step S2, use the vehicle trajectory prediction model to predict the position coordinates and speed changes of the vehicle and surrounding vehicles in the future, and obtain the position and speed data information of the vehicle and surrounding vehicles at each moment in the future time period; S4, based on the position and speed data of the vehicle and surrounding vehicles obtained in step S3, using the vehicle driving risk field model in step 1, respectively calculate the driving risk field of the vehicle in each time period; S5. Determine whether there will be interaction between vehicles based on the inverse of the time taken for the vehicles to collide with each other during normal driving. If there is interaction between vehicles, the overlapping area of ​​multiple driving risk fields is the conflict risk field, and the conflict risk field value is calculated. The conflict risk field values ​​of multiple conflict risk fields in the area where the vehicles that have the interaction are located are added together to obtain the comprehensive conflict risk field value in the area at each time in the changing time period. S6. Outputting the comprehensive conflict risk field value calculated in step S5 in real time, the change of the comprehensive conflict risk field value is the change of the safety risk of vehicles in the area in the future period of time; S7, according to the change of the safety risk of the vehicle in step S6, when the comprehensive conflict risk value of the area increases due to the change of the vehicle's motion state, an early warning is issued; In step S1, the vehicle driving risk field model formula is: In the formula, f i (x, y) is the risk field of vehicle i, M i is the equivalent mass of the vehicle, a is the acceleration of the vehicle, θ is the angle between vehicle i and the surrounding vehicles, p i (x, y, t) is the probability density of the vehicle at (x, y) at time t predicted by the trajectory prediction model, Δt is the time interval between two predictions, γ(t) is the discount function, λ1 and λ2 are the undetermined coefficients 1 and 2, respectively; M i =m i (1.566·10 -14 in 6.687 +0.3345) In the formula, m i is the actual mass of vehicle i, v is the speed of vehicle i; In the formula, 0.5 means that the amount after each half-life is reduced to half of the original amount, t 1 / 2 is the half-life; In step S5, whether there will be interaction between vehicles is determined based on the inverse of the time TTC taken for the vehicles to travel normally until a collision occurs. The interaction determination formula is: In the formula, Interaction is the interaction between vehicles. is the inverse of the TTC of vehicle i, t0 is the prediction start time, and s is the time unit in seconds; The calculation formula of conflict risk field value is: CSF=∫∫f(x,y)g(x,y)...m(x,y)dxdy Where CSF is the conflict risk field value, f(x, y), g(x, y), and m(x, y) are the risk field values ​​of each vehicle that has an interactive impact. The calculation formula for the comprehensive conflict risk field value is: CS Ftotal =∑CSF i Where, CSF total is the comprehensive conflict risk field value in the region, CSF i is the i-th conflict risk field value.

2. The macro-region vehicle early warning method based on the safety risk field according to claim 1 is characterized in that: In step S2, the distance between the vehicle and surrounding vehicles, the geographic location of the vehicle, the size and outline of surrounding vehicles, the acceleration and direction of the vehicle are collected respectively through millimeter wave radar, GPS, laser radar, and combined inertial navigation equipment; the vehicles include the vehicle itself, the front and rear vehicles in the lane of the vehicle itself, and the front and rear vehicles in the left and right lanes.

3. The macro-region vehicle early warning method based on the safety risk field according to claim 1 is characterized in that: In step S3, the vehicle trajectory prediction model uses a GNN model, and the prediction process of the GNN model includes: S31. Preprocess the input data, including: S311 standardizes vehicle status information, wherein the vehicle status information includes position, speed, and acceleration; S312 selects vehicles as nodes, defines edges according to the relative positions between vehicles, and constructs a graph, wherein the feature vector of each vehicle node includes the position, speed, acceleration, direction, and lane of the vehicle; S32. Encoding the state information into a form that can be processed by the GNN model; S33. Perform forward propagation through multiple GNN layers to update the representation of each vehicle node; S34. Use the time module of the recurrent neural network to process the historical trajectory of the vehicle, capture the spatiotemporal characteristics of the vehicle, and the spatiotemporal characteristics include speed and position information to achieve vehicle spatiotemporal relationship modeling; S35.GNN model combines the current state of the vehicle with the learned spatiotemporal features to predict the future position of the vehicle through the fully connected layer; S36.GNN model outputs the predicted position of each vehicle in the future, including a series of coordinate points representing the predicted trajectory.

4. The macro-region vehicle early warning method based on the safety risk field according to claim 3 is characterized in that: In step S33, it includes: S331 Each vehicle node collects feature information of adjacent nodes, where the adjacent nodes represent surrounding vehicles; S332 The current node updates its own state using the collected feature information and its own feature information; S333 uses a multi-layer GNN structure to learn higher-order interactions between nodes and achieve recursive updates of the graph; The structure of multi-layer GNN is: First layer: The information of directly adjacent nodes is aggregated to the current node, and the current node obtains the information of the first-order adjacent nodes; Second layer: The output of the adjacent nodes in the first layer is aggregated with the features of the adjacent nodes again, so that the current node obtains the information of the second-order nodes; More layers: Each additional layer can integrate information from more distant nodes and capture higher-order interactions.

5. The macro-region vehicle early warning method based on the safety risk field according to claim 3 is characterized in that: In step S34, the vehicle history track processing process includes: S341. Input historical trajectory data, including vehicle dynamic information and timestamp information; S342. Convert the trajectory data into a series of time-ordered state vectors, where the state vectors are state information of the vehicle at consecutive time steps, including position and speed; S343. Process the state vector in the embedding layer to reduce the dimension and extract abstract features; S344. Input the embedded features into the RNN model, and the RNN model captures the dynamic changes in the time series data; S345. Each time step of the RNN model outputs a new hidden state to predict the vehicle position at the next time step; S346.RNN model generates an output sequence, which contains predictions of the future position of the vehicle.

6. The macro-region vehicle early warning method based on the safety risk field according to claim 3 is characterized in that: In step S35, the following steps are included: S351. Fusing the spatial features obtained from the GNN layer with the temporal features obtained from the time module of the RNN model, wherein the fusion is achieved by a splicing operation; S352. Through the fully connected layer of the GNN model, the fused features are converted into multiple prediction values ​​suitable for final prediction, and the prediction values ​​directly correspond to the specific representation of the future state of the vehicle, including the future position coordinates, speed and direction, and probability distribution; S353. Use multiple layers of fully connected layers to build a deep network structure, which captures and integrates multi-level abstract features; S354.The last fully connected layer is the output layer, which maps the final high-level features to predicted values, including the location coordinates of the vehicle in the next few time steps.

7. A computer system comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 6.

8. A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that: When the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images