Preventive driving auxiliary method and system based on artificial intelligence

Abstract

The invention relates to the technical field of driving safety, and discloses a preventive driving assistance method and system based on artificial intelligence, and the method comprises the following steps: collecting preventive driving assistance information which comprises driving information, road information and law enforcement resource information; generating structural data by using the collected preventive driving assistance information; and inputting the structural data into a preventive driving model, and respectively outputting a result representing drunk driving judgment, a result representing drunk driving route prediction and a result representing a preventive strategy by the preventive driving model. The driving behavior and the road condition are collected in real time, early warning of potential risks can be achieved, the state and the position of available law enforcement resources are mastered in real time, a law enforcement department can more efficiently dispatch personnel and vehicles, the response time is shortened, the drunk driving intervention efficiency of law enforcement personnel is improved, and the drunk driving intervention effect is improved. And a lot of accident risks caused by drunk driving are effectively reduced.

Description

Technical Field

[0001] The present invention relates to the field of driving safety, and more particularly, to an auxiliary method and system for preventive driving based on artificial intelligence. Background Art

[0002] The threat of drunk driving behavior to public safety has prompted us to explore new technical means for effective prevention. Existing drunk driving detection means mainly rely on random inspections by traffic law enforcement officers and alcohol testing equipment, lacking real-time monitoring and prediction capabilities.

[0003] In adverse weather conditions, the potential risk of drunk driving will increase, and law enforcement officers may not be able to reach the scene in time due to the influence of weather conditions, resulting in the inability to effectively implement intervention and reduce the risk of drunk driving. Summary of the Invention

[0004] The present invention provides an auxiliary method and system for preventive driving based on artificial intelligence to solve the technical problems in related technologies.

[0005] The present invention provides an auxiliary method for preventive driving based on artificial intelligence, including the following steps:

[0006] Step 100, collecting preventive driving assistance information, where the preventive driving assistance information includes driving information, road information, and law enforcement resource information;

[0007] Step 200, generating structured data from the collected preventive driving assistance information;

[0008] Step 300, inputting the structured data into a preventive driving model, and the preventive driving model respectively outputs results representing drunk driving judgment, results representing drunk driving route prediction, and results representing prevention strategies;

[0009] Step 400, conducting preventive driving law enforcement for the driver and their driving vehicle according to the result of drunk driving judgment, and deploying law enforcement resources, including law enforcement officers and law enforcement vehicles, on the road for law enforcement according to the result of drunk driving route prediction, conducting on-site alcohol testing on the target driver, and making corresponding treatments for the driver and their vehicle according to the alcohol test.

[0010] Further, in step 100, the preventive driving assistance information includes the following:

[0011] The driving information includes driver information, driving vehicle information, driving speed information, driving direction information, and driving monitoring video;

[0012] The road information includes weather information, monitoring equipment information on the road, road traffic flow information, and road historical accident information;

[0013] Law enforcement resource information includes the current available numbers of law enforcement officers and vehicles, the locations of law enforcement officers, and the locations of law enforcement vehicles.

[0014] Further, step 200 includes the following steps:

[0015] Step 201, generate first two-dimensional structure data based on driving information. The first two-dimensional structure data includes a first data matrix and a first relationship matrix. A first cell of the first data matrix represents the first one-dimensional structure data of an independent object. The independent objects include drivers, vehicles, roads, and surveillance images. A first cell of the first data matrix only contains the driving information of the independent object it represents;

[0016] The element in the i-th row and j-th column of the first relationship matrix represents the association between the independent objects represented by the i-th cell and the j-th cell of the first data matrix. If there is an association, the value of this element in the first relationship matrix is 1; otherwise, it is 0;

[0017] A driver associates with his corresponding personal information and historical driving records;

[0018] There is a relationship between a driver and the vehicle he drives;

[0019] A vehicle associates with its corresponding vehicle information, the road information it travels on, and the vehicle's traveling speed and direction;

[0020] A surveillance image associates with the surveillance image information in the corresponding driving surveillance video;

[0021] There is a relationship between drivers within the same surveillance image, there is a relationship between vehicles within the same surveillance image, and there is a relationship between the surveillance image and the road in the surveillance image;

[0022] The first one-dimensional structure data includes n data items sorted by time. The t-th data item represents the driving information collected at the t-th moment.

[0023] Further, step 200 also includes the following steps:

[0024] Step 202, generate second two-dimensional structure data from the collected road information. The second two-dimensional structure data includes a second data matrix and a second relationship matrix. A second cell of the second data matrix represents the road information of an independent object. The independent objects include roads, traffic lights, monitoring devices, and weather conditions. A second cell of the second data matrix only contains the road information of the independent object it represents;

[0025] The element in the i-th row and j-th column of the second relationship matrix represents the association between the i-th second unit of the second data matrix and the independent objects represented by the j-th second unit. If there is an association, the value of this element in the second relationship matrix is 1; otherwise, it is 0.

[0026] A road is associated with its geographical location and road status information.

[0027] A monitoring device is associated with its geographical location and the monitored range.

[0028] A traffic signal is associated with its geographical location and the road information it controls.

[0029] A road is associated with the traffic signals and monitoring devices distributed around it.

[0030] There is an association between the traffic signals and monitoring devices on the same road.

[0031] The road, traffic signal, and monitoring device are in the same weather condition.

[0032] Further, the following steps are also included in step 200:

[0033] Step 203: Obtain second one-dimensional structure data based on the sorting of law enforcement resource information. The second one-dimensional structure data includes p data items sorted by time. The t-th data item represents the third two-dimensional structure data generated from the law enforcement resource information collected at the t-th moment.

[0034] The third two-dimensional structure data includes a third data matrix and a third relationship matrix. A third unit of the third data matrix represents the law enforcement resource information of an independent object. The independent objects include law enforcement roads, law enforcement personnel, and law enforcement vehicles. A third unit of the third data matrix only contains the law enforcement resource information of the independent object it represents.

[0035] The element in the i-th row and j-th column of the third relationship matrix represents the association between the i-th third unit of the third data matrix and the independent objects represented by the j-th third unit. If there is an association, the value of this element in the third relationship matrix is 1; otherwise, it is 0.

[0036] A law enforcement road is associated with its geographical location; a law enforcement personnel is associated with its location and law enforcement status; a law enforcement vehicle manages its location and vehicle occupancy information.

[0037] There is a relationship between a law enforcement road and the required law enforcement personnel and law enforcement vehicles; there is a relationship between a law enforcement personnel and the law enforcement vehicle they are in; there is a relationship between law enforcement personnel on the same law enforcement road.

[0038] Further, in step 300, the preventive driving model includes a first intermediate layer, a second intermediate layer, a third intermediate layer, a first feature fusion layer, a first output layer, a second output layer, a fourth intermediate layer, a second feature fusion layer, a fifth intermediate layer, and a third output layer;

[0039] Input the structured data generated from driving information into the first intermediate layer, and output the first intermediate representation data to the second intermediate layer. The second intermediate layer outputs the second intermediate representation data. At the same time, input the structured data generated from road information into the third intermediate layer, and output the third intermediate representation data. The first feature fusion layer inputs the second intermediate representation data and the third intermediate representation data, and outputs the first fusion representation data. The first fusion representation data is input to the first output layer to output the result indicating drunk driving judgment, and the first fusion representation data is input to the second output layer to output the result indicating the predicted drunk driving route;

[0040] Input the structured data generated from law enforcement resource information into the fourth intermediate layer, and output the fourth intermediate representation data to the second feature fusion layer. The second feature fusion layer also inputs the first fusion representation data. The second feature fusion layer outputs the second fusion representation data to the third output layer, and the third output layer outputs the result indicating the preventive strategy.

[0041] Further, in step 300, the result indicating drunk driving judgment refers to whether the driver's state is drunk driving;

[0042] The result indicating the predicted drunk driving route refers to the driving path of the vehicle judged to be drunk driving, and the position that the drunk driving vehicle needs to move to at each future moment;

[0043] The result indicating the preventive strategy refers to preventing the risks brought by drunk driving vehicles and the measures taken by law enforcement officers, including the law enforcement resource scheduling plan and the signal light control plan on the road. The law enforcement resource scheduling plan includes the scheduling behavior of law enforcement officers and the scheduling behavior of law enforcement vehicles. The signal light control plan on the road includes the signal cycle length, the green ratio of each phase, and the phase difference.

[0044] Further, in step 100, it is also necessary to preprocess the collected preventive driving assistance information:

[0045] Step 110, data cleaning:

[0046] Remove noise data: Identify and remove or correct incorrect data, outliers, and duplicate records;

[0047] Fill in missing values: Use statistical methods or model-based methods to predict missing values;

[0048] Unify the format: All data follows the same format standard, including dates and timestamps;

[0049] Step 120, Data Integration:

[0050] Merge data sources: Integrate data from different sources together;

[0051] Resolve conflicts: When obtaining the same type of information from multiple sources and there is data inconsistency, preferentially adopt the most recently updated data;

[0052] Step 130, Data Transformation:

[0053] Normalization: Transform data to the same scale and convert non-numerical data into numerical form;

[0054] Construct calculated fields: Generate new useful information based on existing data;

[0055] Step 140, Data Reduction:

[0056] Dimensionality reduction: Reduce the dimensionality of the dataset through principal component analysis technology while maintaining the key features of the data;

[0057] Aggregation: Summarize data into a higher-level form;

[0058] Sample selection: Select a representative subset from a large amount of data for analysis;

[0059] Step 150, Special processing: For driving surveillance videos, extract key frames, identify specific behaviors, and combine the results with other data for analysis.

[0060] The present invention also provides an artificial-intelligence-based preventive driving assistance system for performing the steps in the aforementioned artificial-intelligence-based preventive driving assistance method, including:

[0061] Data acquisition module: Responsible for acquiring driving information, road information, and law enforcement resource information, including driver information, vehicle information, driving data, road conditions, weather conditions, law enforcement officers, and vehicle deployments;

[0062] Data structuring module: Responsible for generating structured one-dimensional, two-dimensional data matrices, and relationship matrices from the acquired data to represent the association relationships between different types of data;

[0063] Preventive driving model: Composed of multiple neural network layers, including an intermediate layer, a feature fusion layer, and an output layer; Input structured driving data, road data, and law enforcement resource data, and output drunk driving judgment, drunk driving route prediction, and prevention strategies;

[0064] Law enforcement decision-making module: According to the results output by the model, deploy law enforcement resources to conduct on-site inspections and handling of suspected drunk drivers; adjust the signal control strategy on the road and coordinate the signal timing to cooperate with law enforcement operations.

[0065] The present invention also proposes a storage medium storing non-temporary computer-readable instructions for executing the steps in the foregoing auxiliary method for preventive driving based on artificial intelligence.

[0066] The beneficial effects of the present invention are as follows:

[0067] By collecting driving behaviors and road conditions in real time, the present invention can achieve early warning of potential dangers. By mastering the status and location of available law enforcement resources in real time, law enforcement departments can more efficiently dispatch personnel and vehicles, reduce response time, improve the efficiency of law enforcement officers in intervening in drunk driving, and effectively reduce the accident risks caused by many drunk driving incidents. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 is a flowchart of an auxiliary method for preventive driving based on artificial intelligence proposed by the present invention;

[0069] Figure 2 is a structural block diagram of an auxiliary system for preventive driving based on artificial intelligence proposed by the present invention.

[0070] In the figure: 101, data acquisition module; 102, data structuring module; 103, preventive driving model; 104, law enforcement decision-making module. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0071] Now, the subject matter described herein will be discussed with reference to exemplary embodiments. It should be understood that discussing these embodiments is only to enable those skilled in the art to better understand and thus implement the subject matter described herein. Without departing from the protection scope of the content of this specification, the functions and arrangements of the elements discussed can be changed. Each example can omit, substitute, or add various processes or components as needed. Additionally, the features described relative to some examples can also be combined in other examples.

[0072] As Figure 1 shown, an auxiliary method for preventive driving based on artificial intelligence includes the following steps:

[0073] Step 100, collect preventive driving assistance information, where the preventive driving assistance information includes driving information, road information, and law enforcement resource information;

[0074] The driving information includes driver information, driving vehicle information, driving speed information, driving direction information, and driving surveillance video;

[0075] The road information includes weather information, information of monitoring devices on the road, road traffic flow information, and road historical accident information;

[0076] The law enforcement resource information includes the number of currently available law enforcement officers and vehicles, the positions of law enforcement officers, and the positions of law enforcement vehicles;

[0077] Step 200: Generate structured data from the collected driving information, road information, and law enforcement resource information;

[0078] Step 201: Generate a first two-dimensional structured data based on the driving information. The first two-dimensional structured data includes a first data matrix and a first relationship matrix. A first cell in the first data matrix represents the first one-dimensional structured data of an independent object. The independent objects include the driver, the vehicle, the road, and the monitoring image. A first cell in the first data matrix only contains the driving information of the independent object it represents;

[0079] The element in the i-th row and j-th column of the first relationship matrix represents the association between the independent objects represented by the i-th cell and the j-th cell in the first data matrix. If there is an association, the value of this element in the first relationship matrix is 1, otherwise it is 0;

[0080] The driver is associated with his corresponding personal information and historical driving records;

[0081] There is a relationship between the driver and the vehicle he drives;

[0082] The vehicle is associated with its corresponding vehicle information, the road information it travels on, and the vehicle's driving speed and direction;

[0083] The monitoring image is associated with the monitoring image information in the corresponding driving monitoring video;

[0084] There is a relationship between the drivers within the same monitoring image, there is a relationship between the vehicles within the same monitoring image, and there is a relationship between the monitoring image and the road in the monitoring image;

[0085] The first one-dimensional structured data includes n data items sorted by time. The t-th data item represents the driving information collected at the t-th moment;

[0086] Step 202: Generate a second two-dimensional structured data from the collected road information. The second two-dimensional structured data includes a second data matrix and a second relationship matrix. A second cell in the second data matrix represents the road information of an independent object. The independent objects include the road, the traffic signal, the monitoring device, and the weather condition. A second cell in the second data matrix only contains the road information of the independent object it represents;

[0087] The element in the \(i\)-th row and \(j\)-th column of the second relationship matrix represents the association between the \(i\)-th second unit and the independent object represented by the \(j\)-th second unit in the second data matrix. If there is an association, the value of this element in the second relationship matrix is 1; otherwise, it is 0.

[0088] A road is associated with its geographical location and road status information.

[0089] A monitoring device is associated with its geographical location and the monitored range.

[0090] A traffic signal is associated with its geographical location and the road information it controls.

[0091] A road is associated with the traffic signals and monitoring devices distributed around it.

[0092] There is an association between the traffic signals and monitoring devices on the same road.

[0093] The road, traffic signal, and monitoring device are in the same weather condition.

[0094] Step 203: Obtain the second one-dimensional structure data based on the sorting of law enforcement resource information. The second one-dimensional structure data includes \(p\) data items sorted by time. The \(t\)-th data item represents the third two-dimensional structure data generated from the law enforcement resource information collected at the \(t\)-th moment.

[0095] The third two-dimensional structure data includes a third data matrix and a third relationship matrix. A third unit in the third data matrix represents the law enforcement resource information of an independent object. The independent objects include law enforcement roads, law enforcement personnel, and law enforcement vehicles. A third unit in the third data matrix only contains the law enforcement resource information of the independent object it represents.

[0096] The element in the \(i\)-th row and \(j\)-th column of the third relationship matrix represents the association between the \(i\)-th third unit and the independent object represented by the \(j\)-th third unit in the third data matrix. If there is an association, the value of this element in the third relationship matrix is 1; otherwise, it is 0.

[0097] A law enforcement road is associated with its geographical location; a law enforcement personnel is associated with its location and law enforcement status; a law enforcement vehicle manages its location and vehicle occupancy information.

[0098] There is a relationship between a law enforcement road and the required law enforcement personnel and law enforcement vehicles; there is a relationship between a law enforcement personnel and the law enforcement vehicle they are in; there is a relationship between law enforcement personnel on the same law enforcement road.

[0099] Step 300: Input the structure data into the preventive driving model, and the preventive driving model outputs the results representing drunk driving judgment, the results representing drunk driving route prediction, and the results representing preventive strategies respectively.

[0100] The preventive driving model includes a first intermediate layer, a second intermediate layer, a third intermediate layer, a first feature fusion layer, a first output layer, a second output layer, a fourth intermediate layer, a second feature fusion layer, a fifth intermediate layer, and a third output layer;

[0101] Input the structural data generated from driving information into the first intermediate layer, output the first intermediate representation data to the second intermediate layer, the second intermediate layer outputs the second intermediate representation data, and at the same time input the structural data generated from road information into the third intermediate layer, output the third intermediate representation data. The first feature fusion layer inputs the second intermediate representation data and the third intermediate representation data, outputs the first fusion representation data, the first fusion representation data is input to the first output layer to output the result indicating drunk driving judgment, and the first fusion representation data is input to the second output layer to output the result indicating the predicted route of drunk driving;

[0102] Input the structural data generated from law enforcement resource information into the fourth intermediate layer, output the fourth intermediate representation data to the second feature fusion layer, the second feature fusion layer also inputs the first fusion representation data, the second feature fusion layer outputs the second fusion representation data to the third output layer, and the third output layer outputs the result indicating the preventive strategy;

[0103] The result indicating drunk driving judgment refers to whether the driver's state is drunk driving;

[0104] The result indicating the predicted route of drunk driving refers to the driving path of the vehicle judged to be drunk driving, and the position that the drunk driving vehicle needs to move to at each future moment;

[0105] The result indicating the preventive strategy refers to the risks brought by preventing drunk driving vehicles and the measures taken by law enforcement officers, including the law enforcement resource scheduling plan and the signal control plan on the road;

[0106] Among them, the law enforcement resource scheduling plan includes the scheduling behavior of law enforcement officers and the scheduling behavior of law enforcement vehicles, and the signal control plan on the road includes the signal cycle duration, the green time ratio of each phase, and the phase difference;

[0107] In an embodiment of the present invention, the signal cycle duration refers to the total duration required for a complete alternation of the yellow light, red light, and green light in one cycle; the green time ratio of each phase refers to the proportion of the green light duration of each approach lane at an intersection in one cycle; the phase difference refers to the time difference of the green time ratio between adjacent intersection signal lights, which is used to coordinate the signal lights in the area;

[0108] Step 400, conduct preventive driving law enforcement on the driver and their driving vehicle according to the result of drunk driving judgment, and deploy law enforcement resources including law enforcement officers and law enforcement vehicles on the road where the law enforcement is carried out according to the result of the predicted route of drunk driving, conduct on-site alcohol testing on the target driver, and conduct corresponding processing on the driver and their vehicle according to the alcohol test.

[0109] In one embodiment of the present invention, in step 100, it is also necessary to preprocess the collected data:

[0110] 1. Data cleaning: Remove noise data: Identify and remove or correct incorrect data, outliers, and duplicate records.

[0111] For example, for driving speed information, set a reasonable speed range, and speed values outside this range may be regarded as outliers;

[0112] Fill in missing values: Use statistical methods (such as mean, median) or model-based methods to predict missing values.

[0113] For example, if the road traffic flow information for some time periods is missing, interpolation can be performed based on the data of adjacent time periods.

[0114] Unify formats: Ensure that all data follows the same format standard, especially for dates and timestamps. For example, ensure that all time information uses the 24-hour system.

[0115] 2. Data integration:

[0116] Merge data sources: Integrate data from different sources together;

[0117] For example, combine driver information with vehicle information while driving to form a more complete driving record;

[0118] Resolve conflicts: When obtaining the same type of information from multiple sources, data inconsistencies may occur. Rules need to be established to resolve these conflicts, such as preferentially using the most recently updated data.

[0119] 3. Data transformation:

[0120] Normalization: Transform the data to the same scale for subsequent analysis;

[0121] For example, convert all distance units to meters and all speed units to kilometers per hour;

[0122] Encoding: Convert non-numerical data into numerical form;

[0123] For example, convert text descriptions such as "sunny day" and "rainy day" in weather information into digital codes;

[0124] Construct calculation fields: Generate new useful information based on existing data;

[0125] For example, calculate the driving distance based on the driving speed and driving time.

[0126] 4. Data reduction:

[0127] Dimensionality reduction: Reducing the dimensionality of a dataset through techniques such as principal component analysis (PCA) while preserving the key features of the data;

[0128] Aggregation: Summarizing data into a higher-level form to simplify analysis;

[0129] For example, summarizing the vehicle flow information per minute into the vehicle flow information per hour;

[0130] Sample selection: Selecting a representative subset from a large amount of data for analysis to improve efficiency.

[0131] 5. Special processing:

[0132] For driving surveillance videos, it may be necessary to use computer vision techniques to extract key frames or identify specific behaviors (such as signs of drowsy driving), and then use the results together with other data for analysis;

[0133] The law enforcement resource information has strong dynamics and needs to be updated in real time, and may involve privacy protection measures to ensure its use under the premise of legality and compliance.

[0134] In an embodiment of the present invention, the calculation formula of the first intermediate layer is as follows:

[0135] h t,v = ReLU(W hh1 h t-1,v + W xh1 X t,v + b h1 );

[0136] Where h t，v represents the t-th first intermediate representation data of the v-th first unit of the first data matrix, h t-1，v represents the (t - 1)-th first intermediate representation data of the v-th first unit of the first data matrix, X t，v represents the t-th data item of the first one-dimensional structure data corresponding to the v-th first unit of the first data matrix, W hh1 and W xh1 are the first and second weight parameters of the first intermediate layer, b h1 is the bias parameter of the first intermediate layer, and tanh is the hyperbolic tangent function.

[0137] In an embodiment of the present invention, the calculation formula of the second intermediate layer is as follows:

[0138]

[0139] Where k v1The second intermediate representation data representing the v-th first cell of the first data matrix, M1(v) is the set of first cells of the first data matrix associated with the v-th cell of the first data matrix, x v = h n,v , x u = h n,u , h n,v , h n,u respectively represent the n-th first intermediate representation data of the v-th and u-th first cells of the first data matrix, ∈ is a learnable scalar parameter, W k1 represents the two-dimensional recognition weight parameter of the second intermediate layer, and MLP represents a multi-layer perceptron.

[0140] In an embodiment of the present invention, the calculation formula of the third intermediate layer is as follows:

[0141]

[0142] where k v2 represents the third intermediate representation data of the v-th second cell of the second data matrix, d v and d u respectively represent the road information of the independent objects represented by the v-th and u-th cells of the second data matrix, e u，v represents the attention coefficient of the v-th and u-th second cells, e s，v represents the attention coefficient of the v-th and s-th second cells, α u，v represents the normalized attention coefficient of the v-th and u-th second cells, M2(v) represents the set of second cells of the second data matrix associated with the v-th second cell of the second data matrix, exp represents the exponential function with the natural constant as the base, represents the trainable attention vector parameter, CONCAT represents the concatenation operation, T represents the transpose, W k2 represents the two-dimensional recognition weight parameter of the third intermediate layer, and LeakyReLU is the LeakyReLU activation function.

[0143] In an embodiment of the present invention, the calculation formula of the first feature fusion layer is as follows:

[0144]

[0145] where, RH represents the first fusion representation data, θ represents the fusion function (concatenation function or summation function), M3(v) represents the set of all first cells of the first data matrix, M4(v) represents the set of all second cells of the second data matrix, W RH1 represents the summation weight matrix of the first feature fusion layer, and b RH1 represents the summation bias parameter of the first feature fusion layer.

[0146] In one embodiment of the present invention, the calculation formula of the first output layer is as follows:

[0147]

[0148] where y one represents the first output vector, and the i-th component of the first output vector represents the probability that the i-th driver is driving under the influence of alcohol. W y1 is the weight parameter of the first output layer, b y1 is the bias parameter of the first output layer, and σ represents the sigmoid function

[0149] In one embodiment of the present invention, the calculation formula of the second output layer is as follows:

[0150] y two = σ(W y2 RH + b y2 );

[0151] where y two represents the second output vector, and one component of the second output vector represents the longitude and latitude of the node of the driving path of all drunk driving drivers at a future moment. W y2 is the weight parameter of the second output layer, b y2 is the bias parameter of the second output layer, and σ represents the sigmoid function.

[0152] In one embodiment of the present invention, the calculation formula of the fourth intermediate layer is as follows:

[0153] U t,2 = ReLU(H t X t,2 W k3 );

[0154]

[0155] where X t,2 represents the input feature matrix of the t-th data item in the second one-dimensional structure data, and one row vector thereof represents the law enforcement resource information of the independent object represented by a third-order unit of the third-order data matrix of the t-th data item in the second one-dimensional structure data. U t represents the intermediate representation matrix of the t-th data item in the second one-dimensional structure data, and H t represents the intermediate recognition representation, and one row vector thereof represents the fourth intermediate representation data of a third-order unit of the third-order data matrix of the t-th data item in the second one-dimensional structure data, represents the sum of the third-order relationship matrix of the t-th data item in the second one-dimensional structure data and the identity matrix, represents The degree matrix, -0.5 represents the negative 1 / 2 power, W k3 represents the two-dimensional recognition weight parameter of the fourth intermediate layer, and ReLU represents the ReLU activation function.

[0156] In an embodiment of the present invention, the calculation formula of the second feature fusion layer is as follows:

[0157]

[0158] where, RH t，2 represents the t-th second fusion representation data, H t,2,v represents the v-th row vector of the intermediate recognition representation, M RH2 represents the set of the third units of all the third data matrices, θ represents the fusion function (concatenation function or summation function), W RH2 represents the summation weight matrix of the second feature fusion layer, b RH2 represents the summation bias parameter of the second feature fusion layer.

[0159] In an embodiment of the present invention, the calculation formula of the fifth intermediate layer is as follows:

[0160] u (t) =σ(W u F (t) +W u G (t-1) +b u );

[0161] r (t) =σ(W r F (t) +W r G (t-1) +b r );

[0162] C (t) =tanh(W c F (t) +W c r (t) ⊙G (t-1) +b c );

[0163] G (t) =(1 - u (t) )⊙C (t) +u (t) ⊙G (t-1) ;

[0164] where, W u 、W r 、W c represent the first, second, and third weight parameters of the fifth intermediate layer, b u 、br and b c represent the first, second, and third bias parameters of the fifth intermediate layer, ⊙ represents the dot product, and u (t) and r (t) and C (t) represent the first, second, and third intermediate representation data of the fifth intermediate layer respectively, and F (t) =RH t，2 and G (t) and G (t-1) represent the t-th and (t - 1)-th fifth intermediate representation data respectively, where p≥t≥1, p represents the total number of data items of the second one-dimensional structure data, and when t = 1, G (t-1) =F (t) , tanh is the hyperbolic tangent function, and σ represents the sigmoid function.

[0165] In an embodiment of the present invention, the calculation formula of the third output layer is as follows:

[0166] y three =σ(W y3 G (p) +b y3 );

[0167] where y three represents the third output vector, and the f-th component value thereof represents the probability value of the f-th strategy. The strategy with the largest probability value is selected as the output. The strategy group includes all executable strategies. A strategy includes the scheduling methods of law enforcement officers and law enforcement vehicles, and G (p) represents the p-th fifth intermediate representation data, W y3 is the weight parameter of the third output layer, and b y3 is the bias parameter of the third output layer, and σ represents the sigmoid function.

[0168] In an embodiment of the present invention, the steps of training the preventive driving model include:

[0169] Step 101, initialize the parameters of the preventive driving model;

[0170] Step 102, observe the preventive driving assistance information S at time e e , the strategy A executed at time e e , the preventive driving assistance information S at time e + 1 e+1 , and the reward R obtained by executing the strategy A e ; e

[0171] Step 103, then calculate the strategy error:

[0172]

[0173] where δe represents the policy error at time e, γ represents the discount factor, where γ ∈ [0, 1], represents the input S of the preventive driving model P+1 the maximum probability value in the first output vector output when represents the input S of the preventive driving model P when the first output vector output corresponds to policy A e the probability value;

[0174] R e = W1RIS - W2RT - W3RUC - W4PSI;

[0175] where RIS (Risk Intervention Success): represents the number of successful interventions in drunk driving incidents, the more successful the intervention, the higher the reward; RT (Response Time): represents the time required from receiving the report to the arrival of law enforcement officers at the scene, the shorter the response time, the higher the reward, otherwise the reward is reduced; RUC (Resource Utilization Cost): represents the cost of dispatching and using law enforcement resources, including the costs of manpower, vehicles and equipment, and efficient use of resources will result in higher rewards; PSI (Public Safety Improvement): represents the degree of improvement in public safety, which can be measured by indicators such as the reduction in accident rates and the improvement in public satisfaction.

[0176] where W1, W2, W3, and W4 are the weight coefficients of the number of successful interventions in drunk driving incidents, the time required from receiving the report to the arrival of law enforcement officers at the scene, the cost of dispatching and using law enforcement resources, and the degree of improvement in public safety respectively, and the sum of the four is 1, and the default values are 40, 50, 50, and 100.

[0177] Step 104, update the preventive driving model, and the update formula is as follows:

[0178]

[0179] β ∈ [0, 1], β represents the step size of deep learning, ← represents passing the update;

[0180] Step 105, iterate steps 102 - 104 until the preventive driving model converges or the number of iterations reaches the set value. The default value of this value is 50.

[0181] In an embodiment of the present invention, an example of the auxiliary method is as follows:

[0182] I. Collect preventive driving assistance information

[0183] Driving information: Collect the driver's identity information, vehicle model, current driving speed, driving direction, and in-vehicle surveillance video through in-vehicle GPS and cameras.

[0184] Road information: Obtain the current weather conditions using a weather station, understand the traffic flow on the road through traffic surveillance cameras and sensors, and extract accident records that occurred on this section in the past from the historical accident database.

[0185] Law enforcement resource information: Real-time track the locations of available law enforcement officers and law enforcement vehicles, as well as their current status (such as whether they are performing other tasks).

[0186] II. Generate structured data

[0187] Convert the various unstructured and semi-structured data collected above (such as video streams, images, text descriptions) into a structured data format, such as a table in an SQL database.

[0188] For example:

[0189] The Driver info table contains information such as driver ID, name, driver's license number, etc.

[0190] The Vehicle info table records attributes such as vehicle ID, brand, model, etc.

[0191] The Road conditions table covers fields such as section ID, weather conditions, traffic flow level, etc.

[0192] The Law enforcement resources table registers data such as law enforcement officer ID, vehicle ID, current location coordinates, etc.

[0193] III. Input the pre-driving model

[0194] Model training: Use historical data to train the pre-driving model. The pre-driving model can predict the likelihood of drunk driving, possible drunk driving routes, and recommended preventive measures based on the input driving information, road information, and law enforcement resource information.

[0195] Real-time prediction: When new data flows in, the model runs automatically and outputs the following results:

[0196] Drunk driving judgment result: Identify which drivers are suspected of drunk driving;

[0197] Drunk driving route prediction: Predict the possible driving paths of these drivers in the future;

[0198] Prevention strategy recommendation: Propose specific intervention measures for each high-risk driver, such as immediately notifying the nearest law enforcement officers to go for inspection;

[0199] IV. Implement preventive driving law enforcement

[0200] Based on the drunk driving judgment result output by the model, the traffic management department quickly locates the target driver and their vehicle.

[0201] According to the drunk driving route prediction result, reasonably deploy law enforcement officers and law enforcement vehicles on the predicted driving route to ensure that the target vehicle can be intercepted in time.

[0202] When the law enforcement officers arrive at the designated location, conduct an on-site alcohol test on the driver suspected of drunk driving.

[0203] If the alcohol test result is positive, the law enforcement officers will take further actions against the driver according to law, such as temporarily detaining the driver's license and asking them to stop driving, etc.

[0204] Meanwhile, the system will automatically record the entire law enforcement process, including information such as test results, law enforcement time and location, etc., for subsequent analysis and reporting.

[0205] As Figure 2 shown, the present invention also provides an auxiliary system for preventive driving based on artificial intelligence, including:

[0206] Data acquisition module 101: responsible for acquiring driving information, road information and law enforcement resource information, including driver information, vehicle information, driving data, road conditions, weather conditions, law enforcement officers and vehicle deployments;

[0207] Data structuring module 102: responsible for generating structured one-dimensional, two-dimensional data matrices and relationship matrices from the acquired data, representing the association relationships between different types of data;

[0208] Preventive driving model 103: composed of multiple neural network layers, including an intermediate layer, a feature fusion layer and an output layer; input structured driving data, road data and law enforcement resource data, and output drunk driving judgment, drunk driving route prediction and prevention strategies;

[0209] Law enforcement decision-making module 104: according to the results output by the model, deploy law enforcement resources, conduct on-site inspections and handling of drivers suspected of drunk driving; adjust the signal control strategy on the road and coordinate the signal timing to cooperate with the law enforcement action.

[0210] At least one embodiment of the present disclosure provides a storage medium storing non-temporary computer-readable instructions for executing one or more steps in the foregoing auxiliary method for preventive driving based on artificial intelligence.

[0211] A computer program can be stored / distributed on a suitable medium, such as an optical storage medium or a solid-state medium supplied together with other hardware or as part of other hardware, but can also be distributed in other forms, such as via the Internet or other wired or wireless telecommunication systems. Any reference signs in the claims shall not be construed as limiting the scope.

[0212] The above has described the embodiments of the present embodiment, but the present embodiment is not limited to the above specific implementation manners. The above specific implementation manners are merely illustrative rather than restrictive. Under the inspiration of the present embodiment, those of ordinary skill in the art can also make many forms, all of which fall within the protection scope of the present embodiment.

Claims

1. An auxiliary method for preventive driving based on artificial intelligence, characterized in that, Including the following steps: Step 100, collect preventive driving assistance information, which includes driving information, road information, and law enforcement resource information; Step 200, generate structured data from the collected preventive driving assistance information; Step 300, input the structured data into a preventive driving model, and the preventive driving model outputs the results representing drunk driving judgment, the results representing drunk driving route prediction, and the results representing preventive strategies respectively; Step 400, conduct preventive driving law enforcement on the driver and their vehicle according to the result of drunk driving judgment, deploy law enforcement resources on the road where the law enforcement is carried out according to the result of drunk driving route prediction, the law enforcement resources include law enforcement officers and law enforcement vehicles, conduct on-site alcohol tests on the target driver, and conduct corresponding processing on the driver and their vehicle according to the alcohol test.

2. The auxiliary method for preventive driving based on artificial intelligence according to claim 1, characterized in that In step 100, the preventive driving assistance information includes the following: The driving information includes driver information, driving vehicle information, driving speed information, driving direction information, and driving monitoring video; The road information includes weather information, monitoring device information on the road, road traffic flow information, and road historical accident information; The law enforcement resource information includes the current available number of law enforcement officers and vehicles, the positions of law enforcement officers, and the positions of law enforcement vehicles.

3. The auxiliary method for preventive driving based on artificial intelligence according to claim 2, wherein In step 200, the following steps are included: Step 201, generate a first two-dimensional structured data based on the driving information. The first two-dimensional structured data includes a first data matrix and a first relationship matrix. A first cell of the first data matrix represents the first one-dimensional structured data of an independent object. The independent objects include drivers, vehicles, roads, and monitoring images. A first cell of the first data matrix only contains the driving information of the independent object it represents; The element in the i-th row and j-th column of the first relationship matrix represents the association between the independent objects represented by the i-th cell and the j-th cell of the first data matrix. If there is an association, the value of this element in the first relationship matrix is 1, otherwise it is 0; The driver is associated with their corresponding personal information and historical driving records; There is a relationship between the driver and the vehicle they drive; The vehicle is associated with its corresponding vehicle information, the road information it travels on, and the vehicle driving speed and direction; The monitoring image is associated with the monitoring image information in the corresponding driving monitoring video; There is a relationship between the drivers within the same monitoring image, there is a relationship between the vehicles within the same monitoring image, and there is a relationship between the monitoring image and the road in the monitoring image; The first one-dimensional structured data includes n data items sorted by time, and the t-th data item represents the driving information collected at the t-th moment.

4. The auxiliary method for preventive driving based on artificial intelligence according to claim 3, characterized in that In step 200, the following steps are also included: Step 202, generate a second two-dimensional structured data from the collected road information. The second two-dimensional structured data includes a second data matrix and a second relationship matrix. A second cell of the second data matrix represents the road information of an independent object. The independent objects include roads, traffic lights, monitoring devices, and weather conditions. A second cell of the second data matrix only contains the road information of the independent object it represents; The element in the \(i\)-th row and \(j\)-th column of the second relationship matrix represents the association between the \(i\)-th second unit and the \(j\)-th second unit in the second data matrix, which represent independent objects. If there is an association, the value of this element in the second relationship matrix is 1; otherwise, it is 0. A road is associated with its geographical location and road status information. A monitoring device is associated with its geographical location and the monitored range. A signal lamp is associated with its geographical location and the road information it controls. A road is associated with the signal lamps and monitoring devices distributed around it. There is an association between the signal lamps and monitoring devices on the same road. The road, signal lamp, and monitoring device are in the same weather condition.

5. An auxiliary method for preventive driving based on artificial intelligence according to claim 4, characterized in that, Step 200 also includes the following steps: Step 203, obtaining second one-dimensional structure data based on the sorting of law enforcement resource information. The second one-dimensional structure data includes \(p\) data items sorted by time. The \(t\)-th data item represents the third two-dimensional structure data generated from the law enforcement resource information collected at the \(t\)-th moment. The third two-dimensional structure data includes a third data matrix and a third relationship matrix. A third unit in the third data matrix represents the law enforcement resource information of an independent object. The independent objects include law enforcement roads, law enforcement personnel, and law enforcement vehicles. A third unit in the third data matrix only contains the law enforcement resource information of the independent object it represents. The element in the \(i\)-th row and \(j\)-th column of the third relationship matrix represents the association between the \(i\)-th third unit and the \(j\)-th third unit in the third data matrix, which represent independent objects. If there is an association, the value of this element in the third relationship matrix is 1; otherwise, it is 0. A law enforcement road is associated with its geographical location; a law enforcement personnel is associated with its location and law enforcement status; a law enforcement vehicle manages its location and vehicle occupancy information. There is a relationship between a law enforcement road and the required law enforcement personnel and law enforcement vehicles; there is a relationship between a law enforcement personnel and the law enforcement vehicle they are riding in; there is a relationship between law enforcement personnel on the same law enforcement road.

6. The auxiliary method for preventive driving based on artificial intelligence according to claim 5, characterized in that, In step 300, the preventive driving model includes a first intermediate layer, a second intermediate layer, a third intermediate layer, a first feature fusion layer, a first output layer, a second output layer, a fourth intermediate layer, a second feature fusion layer, a fifth intermediate layer, and a third output layer. The structure data generated from driving information is input into the first intermediate layer, and the first intermediate representation data is output to the second intermediate layer. The second intermediate layer outputs the second intermediate representation data. At the same time, the structure data generated from road information is input into the third intermediate layer, and the third intermediate representation data is output. The first feature fusion layer inputs the second intermediate representation data and the third intermediate representation data, and outputs the first fusion representation data. The first fusion representation data is input into the first output layer to output the result representing drunk driving judgment, and the first fusion representation data is input into the second output layer to output the result representing drunk driving route prediction. The structure data generated from law enforcement resource information is input into the fourth intermediate layer, and the fourth intermediate representation data is output to the second feature fusion layer. The second feature fusion layer also inputs the first fusion representation data. The second feature fusion layer outputs the second fusion representation data to the third output layer, and the third output layer outputs the result representing the preventive strategy.

7. An auxiliary method for preventive driving based on artificial intelligence according to claim 6, characterized in that, In step 300, the result representing drunk driving judgment refers to whether the driver's state is drunk driving. The result indicating the prediction of drunk driving routes refers to the driving paths of vehicles that are predicted to be driving under the influence of alcohol, and the positions that the vehicles driving under the influence of alcohol need to move to at each future moment; The result indicating the prevention strategy refers to the risks brought by preventing vehicles driving under the influence of alcohol, and the measures taken by law enforcement officers, including law enforcement resource scheduling plans and signal control plans on the road. Among them, the law enforcement resource scheduling plan includes the scheduling behaviors of law enforcement officers and law enforcement vehicles, and the signal control plan on the road includes signal cycle duration, green time ratio per phase, and phase difference.

8. An auxiliary method for preventive driving based on artificial intelligence according to claim 7, characterized in that, In step 100, it is also necessary to preprocess the collected preventive driving assistance information: Step 110, data cleaning: Removing noise data: Identifying and removing or correcting incorrect data, outliers, and duplicate records; Filling in missing values: Using statistical methods or model-based methods to predict missing values; Unifying formats: All data follows the same format standard, including dates and timestamps; Step 120, data integration: Merging data sources: Integrating data from different sources together; Resolving conflicts: When there is data inconsistency when obtaining the same type of information from multiple sources, preferentially adopt the most recently updated data; Step 130, data transformation: Normalization: Transforming data to the same scale and converting non-numerical data into numerical form; Constructing calculated fields: Generating new useful information based on existing data; Step 140, data reduction: Dimensionality reduction: Reducing the dimensionality of the dataset through principal component analysis technology while maintaining the key features of the data; Aggregation: Summarizing data into a higher-level form; Sample selection: Selecting a representative subset from a large amount of data for analysis; Step 150, special processing: For driving surveillance videos, extracting key frames, identifying specific behaviors, and combining the results with other data for analysis.

9. An auxiliary system for preventive driving based on artificial intelligence, characterized in that, For performing the steps in an assisted method for preventive driving based on artificial intelligence as described in any one of claims 1-8, including: Data acquisition module: Responsible for acquiring driving information, road information, and law enforcement resource information, including driver information, vehicle information, driving data, road conditions, weather conditions, law enforcement officers, and vehicle deployments; Data structuring module: Responsible for generating structured one-dimensional, two-dimensional data matrices and relationship matrices from the collected data, representing the association relationships between different types of data; Preventive driving model: Composed of multiple neural network layers, including intermediate layers, feature fusion layers, and output layers; Inputting structured driving data, road data, and law enforcement resource data, and outputting drunk driving judgments, drunk driving route predictions, and prevention strategies; Law enforcement decision-making module: According to the results output by the model, deploying law enforcement resources, conducting on-site inspections and handling of suspected drunk driving drivers; Adjusting the signal control strategy on the road and coordinating the signal timing to cooperate with law enforcement operations.

10. A storage medium, characterized in that, Stored with non-transitory computer-readable instructions for performing the steps in an assisted method for preventive driving based on artificial intelligence as described in any one of claims 1-8.

Images