Method and system for judging fatigue driving based on multi-dimensional data

Abstract

The invention discloses a method and a system for judging fatigue driving based on multi-dimensional data. The method comprises the following steps: acquiring perception data in real time; defining a continuous driving record table for storing an analysis result of continuous vehicle driving of the driver; for the driving vehicle and the driver whose continuous driving time exceeds a first preset time, preliminarily determining that the driving vehicle and the driver are suspected fatigue driving, and obtaining attribute information of the suspected fatigue driving vehicle and the driver; driving state information of a driver is continuously collected and counted, and whether the fatigue driving condition really exists or not is judged and marked; for suspected fatigue drivers, physiological fatigue driving judgment needs to be carried out; the scheme does not need to depend on vehicle-mounted professional equipment, can effectively cover the stock vehicle market, and provides real-time early warning and monitoring capabilities for a traffic management department, so that the traffic accident risk caused by fatigue driving is remarkably reduced.

Description

Technical Field

[0001] The present invention relates to the field of safe driving, and in particular to a method and system for determining fatigue driving based on multidimensional data. Background Art

[0002] Fatigue driving refers to a situation where the driver drives for more than eight hours a day, or engages in other labor that consumes too much physical energy or lacks sleep, resulting in drowsiness and weakness in the limbs while driving, and is unable to promptly detect and accurately deal with road traffic conditions.

[0003] Although the maturity of biometric technology has spawned a series of patented technologies for determining fatigue driving status, these technologies generally rely on installing professional equipment on the vehicle for technical identification. However, for owners of old vehicles or those who consider costs, the willingness to install such new professional equipment is low, resulting in such patented technologies being only applicable to single-unit scenarios and having extremely limited application in the stock vehicle market. In addition, it is difficult for relevant departments to achieve early warning and monitoring through such technologies, and it is impossible to eliminate potential accidents in the bud.

[0004] With the rapid development of video surveillance technology and big data technology, the coverage of face capture equipment and vehicle capture equipment in cities has become increasingly comprehensive, and the quality of the perception data captured has been significantly improved. At the same time, the technical investigation data and network security data mastered by relevant departments have become richer.

[0005] Therefore, how to judge fatigue driving based on existing public data and issue early warnings is an urgent problem that needs to be solved. Summary of the invention

[0006] To this end, it is necessary to provide a method to determine fatigue driving based on the existing computing and storage resources of relevant departments in various places, and to make full use of multi-dimensional data such as perception data, technical investigation data, and network security data.

[0007] To achieve the above object, the inventor provides a method for determining fatigue driving based on multidimensional data, comprising the following steps: S101, real-time acquisition of perception data; S102, defining a continuous driving record table for storing analysis results of the driver's continuous driving of the vehicle; S103, for a driving vehicle and a driver whose continuous driving time exceeds a first preset time, preliminarily determining that the driving is suspected of fatigue driving, and obtaining attribute information of the vehicle and the driver suspected of fatigue driving; S104, continuously collecting and counting the driving status information of the driver, and after verification, marking whether fatigue driving actually occurs; S105: For suspected fatigue drivers, if the continuous driving time is T continuousIf the second preset time is exceeded and the last capture time is between 5:00 and 9:00 in the morning, a determination of physiological fatigue driving needs to be made; if the continuous driving time exceeds the third preset time, it is directly determined as continuous fatigue driving.

[0008] As a preferred embodiment of the present invention, in step S101, the real-time acquisition of perception data includes acquiring face capture data, vehicle capture data, and main driver capture data.

[0009] As a preferred embodiment of the present invention, step S102 further includes: Using a stream processing framework to analyze the newly acquired main driver capture data to determine whether there is a record in the continuous driving record table of the in-memory database for the vehicle being driven; If there is no record, use the license plate number of the vehicle being driven to query in the vehicle capture data whether there is capture data of this vehicle within T minutes. Let the current time be t current , the first query time window is t current - T min , t current , if there is, continue to query T minutes further back, that is, the next query time window is t current – 2*T min , t current - T min , and so on in a loop until there is no capture data of this vehicle within a T-minute time period. Take the last capture data as the earliest capture data of this vehicle on the day and record it in the continuous driving record table of the in-memory database; If there is a record, obtain the last capture time recorded t last , and within this time and this capture time t current , that is t last , t current , check whether there is capture data of this vehicle every T minutes. If there is, update the continuous driving record table, update the last capture time to this capture time t last , and calculate the continuous driving time. If there is an interruption, that is t current - (n + 1) * T min , t current - n* T min There is a capture record within the time period, and the capture time is t newstart , while t current - (n + 2) * T min , t current - (n + 1)*T min If there is no capture record within the time period, it is considered that within this time period, the driver has rested for more than T minutes and there is no continuous driving situation. At this time, it is necessary to update the earliest capture time to t newstart , and update the most recent capture time to t last , and calculate the continuous driving time T continuous , where n represents a natural number, t current - (n + 1) * T min , t current - n * T min represents n + 1 T's backward from t current going back min time period.

[0010] As a preferred embodiment of the present invention, in step S103, obtaining the attribute information of the suspected fatigue driving vehicle and driver includes vehicle registration information, driver information, driver's illegal driving record, and driver's driving accident record; through the keyword fields in the attribute information for association and matching, the above multi-source data is integrated to form a comprehensive data set.

[0011] As a preferred embodiment of the present invention, in step S104, continuously collect and statistically analyze the driving state information of the driver. After verification, the label indicating whether there is actually fatigue driving includes: y = 0 indicates error, that is, the driver does not have fatigue driving; y = 1 indicates correct, that is, the driver has fatigue driving; Among them, y represents the label type of fatigue driving.

[0012] As a preferred embodiment of the present invention, step S104 further includes data cleaning and feature processing steps: Clean the data and process the features of the labeled dataset, controlling the ratio of the number of data with label types 1 and 0 to be 8:2; Perform vectorization on the cleaned data, represented as a feature vector x = x 1 , x 2 , x 3 , x 4 , x 5 , x 6 , x 7 , x 8 , and divide the fields into driver-related feature fields and vehicle-related feature fields; Among them, the driver-related feature fields are extracted from the driver information, including: Driving age: x 1 = Current date - First license date; Driver's license status code: Numerically encode different driver's license statuses, x 2 = {0: Normal; 1: Revoked; 2: Suspended}; Number of violations in the past year: x 3 = Number of traffic violations in the past year; Average severity of accidents in the past year: After quantitatively scoring the severity of accidents based on the accident type and loss, take the average, , where, Represents the severity score of the i-th accident; Among them, the vehicle-related feature fields are extracted from the vehicle registration information, including: Vehicle age: x 5 = Current date - Vehicle registration date; Vehicle type code: Numerically encode different vehicle types, x 6 = {0: Sedan; 1: Truck; 2: Bus}; Vehicle use nature: x 7 = {0: Operating; 1: Non-operating}; Number of overdue annual inspections: x 8 = Number of times of overdue annual inspections within the past preset time; Using the random sampling method, 80% of the labeled dataset is selected as the training set and 20% as the test set. The ratio of the number of label types of 0 and 1 in the training set and the test set is the same. The training set is trained using the random forest method. The initial value of the number of decision trees in the random forest model is set to 30, the initial value of the maximum depth parameter of the decision tree is set to 5, and the initial value of the minimum number of samples required for node splitting is set to 1. The accuracy and F1-score metrics are calculated using the test set to confirm the model effect. Through the parameter tuning method of random search, during several iterative training processes, the parameters are continuously adjusted for data training; The random forest model is trained. For the suspected fatigue driving data collected in real time, the relevant information of the driver and the vehicle is queried, and the random forest model is called for prediction, and the probability of suspected fatigue driving is output, denoted as V feature , V feature which is a probability value between 0 and 1, representing the probability that the driver is in a fatigue driving state.

[0013] As a preferred embodiment of the present invention, in step S105, the determination of physiological fatigue driving includes: using the driver's ID card information to query the following data during the period from 23:00 the previous day to 5:00 the next morning of the driver: A. Whether there is face capture data; B. Obtain the affiliated mobile phone number, query the mobile phone call data, and return the last call time; C. Obtain the affiliated mobile phone number, query the last use time of the mobile phone APP; D. Query the transportation means ride information; E. Query the Internet access end time in the Internet cafe; F. Query the venue registration time; Introduce the nighttime activity index N for determining physiological fatigue. The expression is:

[0014] where δ is the indicator function, ∑δ1 represents the total number of capture events during the statistical time period, ∑δ2 represents the total number of registrations during the statistical time period, and ∑δ3 represents the total number of calls and mobile phone APP uses during the statistical time period; Obtain the historical benchmark value λ of the nighttime activity index N through historical data, and determine the nighttime activity index threshold N threshold , and through hypothesis testing to judge whether the driver is in a fatigue risk state. The null hypothesis H0: N ≤ N threshold represents that the driver is in a normal rest state, and the alternative hypothesis H1: N > N thresholdIndicates that the driver has a fatigue risk; P(N≤N threshold ) is calculated through the cumulative distribution function of the Poisson distribution, and the expression is:

[0015] Among them, P(N≤N threshold ) is the probability of N≤N threshold , that is, the probability that the driver is in a normal rest state; k is the specific number of occurrences; e is the natural constant; k! is the factorial of k; Then the calculation expression of the probability P fatigue that the driver is at fatigue risk is:

[0016] As a preferred embodiment of the present invention, for the determination of physiological fatigue driving, it further includes step S106, which uses multi-factor weights to determine. Let the physiological fatigue accumulation factor be F accumulation , and its calculation expression is:

[0017] Among them, is the weight coefficient; If the continuous driving time T continuous exceeds two hours and F accumulation >1, it is determined as physiological fatigue driving.

[0018] As a preferred embodiment of the present invention, it further includes a trigger warning step. For the records of physiological fatigue driving and continuous fatigue driving, they are displayed on the display interface. After review, a notice announcement, trigger a text message and / or a phone voice reminder the driver to take a rest.

[0019] To achieve the above object, the inventor also provides a system for determining fatigue driving based on multi-dimensional data, including: A unified docking module for obtaining multi-dimensional perception data; A big data storage module for storing the accessed multi-dimensional perception data; A big data analysis module for performing data analysis on the accessed perception data to judge physiological fatigue driving and continuous fatigue driving; An alarm reminder module for displaying the information of physiological fatigue driving and continuous fatigue driving and triggering an alarm.

[0020] Different from the prior art, the beneficial effects achieved by the above technical solutions are: (1) This method and system do not need to rely on professional in-vehicle equipment, can effectively cover the existing vehicle market, and provide real-time early warning and monitoring capabilities for relevant departments, thus significantly reducing the risk of traffic accidents caused by fatigue driving. (2) This method and system can be completely based on the existing data of vehicle and driver capture, monitoring, etc. by relevant departments, reuse the existing computing and storage resources, and simply use software programs to determine fatigue driving. (3) This method and system can comprehensively refer to multi-dimensional data, conduct comprehensive determination and analysis of fatigue driving, and make the analysis results more accurate. Description of the Drawings

[0021] Figure 1 It is the flowchart of the method described in the specific implementation manner; Figure 2 It is the system framework diagram described in the specific implementation manner. Specific Implementation Manner

[0022] In order to elaborate in detail the technical content, structural features, achieved objectives and effects of the technical solution, the following is described in detail in conjunction with specific embodiments and with reference to the accompanying drawings.

[0023] In the following embodiments, physiological fatigue driving means: continuously driving for two hours between 5:00 and 9:00 in the morning, and resting less than six hours between 23:00 the previous day and 5:00 the next day; continuous fatigue driving means: the continuous fatigue driving defined in the "Regulations on Road Traffic Management" is driving for more than eight hours a day. In actual situations, driving continuously for four hours will cause fatigue conditions such as drowsiness and weakness in the limbs.

[0024] As Figure 1 shown, this embodiment provides a method for determining fatigue driving based on multi-dimensional data, specifically including the following steps: S101, obtain perception data in real time, including: Face capture data: capture time, capture location, real-name information of the captured face, and the real-name information includes ID number, name, etc.; after such data is captured in real time by a face capture device, the face photo is structurally analyzed, and after real-name authentication is performed in combination with the personnel real-name information, it is formed.

[0025] Vehicle capture data: capture time, capture location, captured vehicle license plate number, captured vehicle type, such as: small car, large bus, truck, etc.; after such data is captured in real time by a vehicle capture device, the vehicle photo is structurally analyzed to obtain the vehicle license plate number and vehicle type.

[0026] Driver capture data: capture time, capture location, license plate number of the captured vehicle, type of the captured vehicle, such as small car, large bus, truck, etc.; ID number and name of the main driver, etc.; Compared with ordinary vehicle capture data, driver capture data relies on higher-precision capture devices and is installed in the vehicle driving direction for capturing the front of the vehicle. After capturing the photo, the vehicle photo is structurally analyzed to obtain the license plate number and vehicle type of the vehicle, and the face photo of the main driver is structurally analyzed, and real-name authentication is performed in combination with the real-name information of the person; The above data belongs to the basic business data maintained by relevant departments and is constructed, maintained, and docked in accordance with the relevant specifications of GA / T 1400. After the system obtains the perception data, it is stored in the local big data warehouse.

[0027] S102, define a table as the continuous driving record table, which is used to store the analysis results of the driver's continuous driving of the vehicle, including the main driver's ID number, driving license plate number, earliest capture time, last capture time, calculation of continuous driving time and other main fields. The value of the continuous driving time is the time difference between the earliest capture time and the last capture time. The continuous driving record table is stored in the in-memory database to achieve the effect of low latency and high concurrency.

[0028] Step S102 also includes: using a stream processing framework to analyze the newly accessed driver capture data to determine whether there is a record in the continuous driving record table of the vehicle being driven in the in-memory database: If there is no record, use the license plate number of the vehicle being driven to query whether there is capture data of this vehicle within T minutes in the vehicle capture data. In this embodiment, the value of T minutes can be 20 minutes. Let the current time be t current , the first query time window is t current - 20 min , t current . If there is, continue to query 20 minutes further back, that is, the next query time window is t current - 40 min , t current - 20 min , and so on in a loop until there is no capture data of this vehicle in a certain 20-minute time period. Take the last capture data as the earliest capture data of this vehicle on the same day and record it in the continuous driving record table of the in-memory database; If there is a record, obtain the last capture time of the record t last , at this time and this capture time tcurrent , that is t last , t current , within the time period, check if there is capture data of this vehicle every 20 minutes. If there is data for every interval, update the continuous driving record form, and update the most recent capture time to this capture time t last , and calculate the continuous driving time; if there is an interruption, that is t current - ( n+1 ) * 20min , t current - n * 20min , there is a capture record within the time period, and the capture time is t newstart , while t current - (n + 2) * 20 min , t current - (n + 1) * 20 min , there is no capture record within the time period, it is considered that the driver has rested for more than 20 minutes during this time period, and there is no continuous driving situation. At this time, it is necessary to update the earliest capture time to t newstart , and update the most recent capture time to t last , and calculate the continuous driving time T continuous ; where n is a natural number, t current - (n + 1) * 20 min , t current - n * 20 min represents going back n + 1 intervals of 20 t current backwards from min the time period.

[0029] S103. For the continuous driving time T continuous that exceeds the first preset time, that is, two hours of driving vehicles and drivers, it is initially determined as suspected of fatigue driving. The system is docked with the business platforms of relevant departments through the synchronization interface method to obtain the following data information of the suspected fatigue driving vehicles and drivers.

[0030] Vehicle registration information: including vehicle license plate number, such as license plate number; vehicle type, such as: small car, large bus, truck, etc.; vehicle brand, vehicle model, vehicle color, vehicle identification code VIN, engine number, vehicle registration date, vehicle use nature, such as: operating, non-operating; vehicle annual inspection records, etc.

[0031] Driver information: including driver's name, gender, date of birth, ID number, contact information, such as: phone number, address, etc.; driver's license number, driver's license type, such as C1, B2, A1, etc.; driver's license validity period, including start date and end date; driver's license issuing authority, permitted driving models, date of first obtaining the license, driver's license status, such as normal, revoked, cancelled, suspended, etc.; driver's license scoring situation, that is, the cumulative score within the current scoring cycle; driver's license verification records, such as verification date, verification result, etc.

[0032] Driver's illegal driving records: including time of illegal occurrence, location of illegal occurrence, type of illegal, such as speeding, running a red light, drunk driving, etc.; illegal handling status, illegal punishment result, including fine amount, points deduction situation, etc.

[0033] Driver's driving accident records: including time of accident occurrence, location of accident occurrence, type of accident, such as rear-end collision, collision, rollover, etc.; accident liability determination situation, accident handling result, such as compensation amount, liability ratio, etc.; personnel injury and death situation, vehicle damage situation, such as minor damage, serious damage, scrapped, etc.

[0034] By associating and matching through keyword fields such as ID number and license plate number, the above multi-source data is integrated to form a comprehensive data set to ensure data consistency and accuracy.

[0035] S104, continuously collect and count the driving status information of drivers, and after manual verification, label whether there is actual fatigue driving. Here, the label is represented by y for indication.

[0036] y = 0 indicates error, that is, the driver does not have fatigue driving; y = 1 indicates correct, that is, the driver has fatigue driving.

[0037] Perform data cleaning and feature processing on the labeled data set, and control the data quantity ratio of labels of type 1 and 0 to be 8:2.

[0038] Perform vectorization processing on the cleaned data, and represent it as a feature vector x = x 1 , x 2 ,x 3 , x 4 , x 5 , x 6 , x 7 , x 8 , each field is divided into two categories: One category is: driver-related features, extracted from driver information, including: (1) Driving age: x 1 = Current date - Date of first obtaining a driver's license; (2) Driver's license status code: Numerically encode different driver's license statuses, x 2 = {0: Normal; 1: Revoked; 2: Suspended}; (3) Number of violations in the past year: x 3 = Number of traffic violations in the past year; (4) Average severity of accidents in the past year: After quantitatively scoring the severity of accidents according to the type of accident and the loss situation, take the average value, , where, represents the severity score of the i-th accident; The other category is: vehicle-related features, extracted from vehicle registration information, including: (1) Vehicle age: x 5 = Current date - Vehicle registration date; (2) Vehicle type code: Numerically encode different vehicle types, x 6 = {0: Sedan; 1: Truck; 2: Bus}; (3) Vehicle use nature: x 7 = {0: Operating; 1: Non-operating}; (4) Number of overdue annual inspections: x 8 = Number of times of overdue annual inspections within a certain period in the past.

[0039] Using the random sampling method, 80% of the labeled dataset is selected as the training set and 20% as the test set. It is necessary to ensure that the ratio of the number of label types of 0 and 1 in the training set and the test set is the same to ensure that the training set and the test set are representative in terms of data distribution and avoid the impact of data deviation on the model results. Use the random forest method to train the training set. Set the initial value of the number of decision trees in the random forest model to 30, the initial value of the maximum depth parameter of the decision tree to 5, and the initial value of the minimum number of samples required for node splitting to 1. Calculate metrics such as accuracy and F1 score using the test set to confirm the model effect. Through the hyperparameter tuning method of random search, continuously adjust the parameters during multiple iterative training processes to enable the model to better fit the training data. At the same time, as data accumulates and is analyzed, retrain and adjust the model regularly.

[0040] Finally, a random forest model is trained. The program queries the relevant information of the driver and the vehicle for the suspected fatigue driving data collected in real time, calls the model for prediction, and outputs the probability of suspected fatigue driving, denoted as V feature , V feature which is a probability value between 0 and 1, representing the probability that the driver is in a fatigued driving state.

[0041] S105. For suspected fatigue driving personnel, if the continuous driving time T continuous exceeds the second preset time, that is, two hours, and the last capture time is between 5:00 and 9:00 in the morning, it is necessary to determine physiological fatigue. Use the driver's ID number to query during the period from 23:00 the previous day to 5:00 the next day for the driver: A. Whether there is face capture data; B. Obtain the affiliated mobile phone number, query the mobile phone call data, and return the last call time; C. Obtain the affiliated mobile phone number, query the last usage time of mobile phone apps such as online shopping, WeChat, QQ, and Douyin; D. Query the travel information of transportation means such as airplanes and high-speed rails; E. Query the end time of Internet access in Internet cafes; F. Query the registration time of places such as scenic spots, hotels, etc.; The data of A above has been obtained in step S1. The data of B to F belongs to the basic business data maintained by relevant departments and is connected through a synchronous interface method.

[0042] Introduce the nighttime activity index N for determining physiological fatigue. The expression is:

[0043] where δ are all indicator functions; δ1 represents that if a capture event occurs within this time period (from 23:00 to 5:00 the next day), the value of δ is 1; if no capture event occurs, the value of δ is 0; ∑δ1 represents the total number of capture events counted within this time period. δ2 represents that if there is registration information of a specific place not related to rest, the value of δ is 1, otherwise it is 0; ∑δ2 represents the total number of registrations counted within this time period. δ3 represents that if there is a mobile phone call record or app usage record, the value of δ is 1, otherwise it is 0; ∑δ3 represents the total number of calls and app usage counted within this time period.

[0044] The historical reference value λ of the night activity index N is obtained through a large amount of historical data, and the night activity index threshold N is determined. threshold . Whether the driver is in a fatigue risk state is judged through a hypothesis test. The null hypothesis H0: N ≤ N threshold indicates that the driver is in a normal rest state, and the alternative hypothesis H1: N > N threshold indicates that the driver has a fatigue risk.

[0045] P(N ≤ N threshold ) can be calculated through the cumulative distribution function of the Poisson distribution, and the expression is:

[0046] where: P(N ≤ N threshold ) is the probability of N ≤ N threshold , that is, the probability that the driver is in a normal rest state; N is a random variable representing the number of times an event occurs, which is the value of the night activity index here N ; k is the specific number of occurrences, k = 0, 1, 2, ⋯; λ is the parameter of the Poisson distribution, which represents the average number of events occurring per unit time (or space), and here it is the historical reference value of the night activity index obtained through a large amount of historical data; e is the natural constant, approximately equal to 2.71828; k! is the factorial of k, that is, k! = k × (k - 1) × ⋯ × 1, and it is stipulated that 0! = 1.

[0047] Then the probability P fatigue that the driver is in a fatigue risk state is calculated by the following expression: , S106. For the determination of physiological fatigue driving, multi-factor weights are used. Let the physiological fatigue accumulation factor be F accumulation , and its calculation expression is:

[0048] where is the weight coefficient; If the continuous driving time T continuous exceeds two hours and F accumulation > 1, it is determined as physiological fatigue driving; S107. If the continuous driving time T continuous exceeds the third preset time, that is, four hours, it is directly determined as continuous fatigue driving.

[0049] S108. Trigger an alarm. Records of physiological fatigue driving and continuous fatigue driving are displayed on a unified interface. After being reviewed by a dedicated person arranged by the relevant department, a notice and announcement are made; at the same time, it also supports automatically triggering text messages and phone voice reminders to alert the driver to rest, taking preventive measures.

[0050] As Figure 2 shown, this embodiment also provides a system for determining fatigue driving based on multi-dimensional data, including: A unified docking module, mainly used to obtain multi-dimensional perception data through the GA / T 1400 protocol, including face capture data, vehicle capture data, and main driver capture data; through the synchronous interface method, it docks with the business systems of external relevant departments and traffic business systems for real-time interface queries; A big data storage module, mainly used to store the accessed multi-dimensional perception data, supporting PB-level data storage and second-level queries; A big data analysis module, mainly used to analyze the accessed perception data, and judge physiological fatigue driving and persistent fatigue driving according to big data analysis technology; An alarm reminder module, which displays information on physiological fatigue driving and persistent fatigue driving, and is responsible for interacting with external alarm systems, such as the SMS center and the outbound call center, for automated alarms.

[0051] This system is also used to run and complete the method steps in any of the above embodiments.

[0052] The key points of the method and system of the above embodiments lie in performing multi-source fusion analysis on face capture data, vehicle capture data, and driver behavior data; predicting and analyzing using a random forest model by analyzing the basic information of the driver and the vehicle being driven; inferring whether the driver has not rested for a long time by analyzing the driver's behavior data such as phone calls, Internet access, shopping, and mobile phone use; designing a real-time data processing and warning system that can quickly analyze data and issue a warning of fatigue driving; and being able to be linked with relevant departments for convenient and timely notification and announcement.

[0053] It should be noted that although the above embodiments have been described in this article, the patent protection scope of the present invention is not limited thereby. Therefore, based on the innovative concept of the present invention, any changes and modifications made to the embodiments described in this article, or equivalent structural or equivalent process transformations made using the content of the specification and drawings of the present invention, and directly or indirectly applying the above technical solutions to other related technical fields, are all included in the patent protection scope of the present invention.

Claims

1. A method for determining fatigue driving based on multi-dimensional data, characterized in that, It includes the following steps: S101, obtaining perception data in real time; S102, defining a continuous driving record form for storing the analysis results of the driver's continuous driving of the vehicle; S103, for the vehicle and driver whose continuous driving time exceeds the first preset time, initially determining them as suspected of fatigue driving, and obtaining the attribute information of the suspected fatigue driving vehicle and driver; S104, continuously collecting and statistically analyzing the driving status information of the driver, and after verification, tagging whether there is a real fatigue driving situation; S105. For a driver suspected of fatigue driving, if the continuous driving time T continuous exceeds the second preset time and the most recent capture time is between 5:00 and 9:00 in the morning, a physiological fatigue driving determination is required; if the continuous driving time exceeds the third preset time, it is directly determined as continuous fatigue driving.

2. The method for determining fatigue driving based on multi-dimensional data according to claim 1, wherein: In step S101, obtaining perception data in real time includes obtaining face capture data, vehicle capture data, and main driver capture data.

3. The method for determining fatigue driving based on multi-dimensional data according to claim 2, wherein Step S102 further includes: Using a stream processing framework to analyze the newly obtained main driver capture data, and determining whether there is a record in the continuous driving record form of the in-memory database of the vehicle being driven; If there is no record, use the license plate number of the vehicle being driven to query the vehicle capture data to check if there is capture data of this vehicle within T minutes. Let the current time be t current , the first query time window is t current - T min , t current . If there is, continue to query by pushing back T minutes further, that is, the next query time window is t current – 2*T min , t current - T min . Repeat this cycle until there is no capture data of this vehicle within a T-minute time period. Take the last capture data as the earliest capture data of this vehicle on the same day and record it in the continuous driving record table of the in-memory database; If there is a record, obtain the most recent capture time of the record t last , and within this time and the current capture time t current , that is t last , t current , check whether there is capture data of this vehicle every T minutes. If there is, update the continuous driving record form and update the most recent capture time to the current capture time t last , and calculate the continuous driving time. If there is an interruption, that is t current - (n + 1) * T min , t current - n * T min , if there is a capture record within the time period, and the capture time is t newstart , while t current - (n + 2) * T min , t current - (n + 1)*T min , there is no capture record, it is considered that within this time period, the driver has rested for more than T minutes and there is no continuous driving situation. At this time, it is necessary to update the earliest capture time to t newstart , and update the most recent capture time to t last , and calculate the continuous driving time T continuous , where n represents a natural number, t current - (n + 1) * T min , t current - n * T min represents a time period that goes back n + 1 Ts from t current going back min time period.

4. The method for determining fatigue driving based on multi-dimensional data according to claim 1, wherein: In step S103, obtaining the attribute information of the suspected fatigue driving vehicle and driver includes vehicle registration information, driver information, driver's illegal driving records, and driver's driving accident records; Through association and matching using the key fields in the attribute information, the above multi-source data is integrated to form a comprehensive data set.

5. The method for determining fatigue driving based on multi-dimensional data according to claim 1, wherein: In step S104, continuously collecting and statistically analyzing the driving status information of the driver, and after verification, tagging whether there is a real fatigue driving situation includes: y = 0 indicates an error, i.e., the driver does not have a fatigue driving situation; y = 1 indicates correct, that is, the driver is suffering from fatigue driving; Where y represents the label type of the fatigue driving situation.

6. The method for determining fatigue driving based on multi-dimensional data according to claim 5, wherein: Step S104 further includes data cleaning and feature processing steps: Performing data cleaning and feature processing on the tagged data set, and controlling the ratio of the number of data with label types 1 and 0 to be 8:2; Vectorize the data after cleaning, represented as a feature vector x = x 1 , x 2 , x 3 , x 4 , x 5 , x 6 , x 7 , x 8 , and divide the fields into driver-related feature fields and vehicle-related feature fields; Among them, the driver-related feature fields are extracted from the driver information, including: Driving experience: x 1 = Current date - Date of initial license issuance; Driver's license status code: Numerically encode different driver's license statuses. x 2 = {0: Normal; 1: Revoked; 2: Suspended}; Number of violations in the past year: x 3 = Number of traffic violations in the past year; Average severity of accidents in the past year: After quantifying and scoring the severity of accidents based on accident types and losses, the average value is taken, , where represents the severity score of the i-th accident; Among them, the vehicle-related feature fields are extracted from the vehicle registration information, including: Vehicle service life: x 5 = Current date - Vehicle registration date; Vehicle type coding: Numerically code different vehicle types, x 6 = {0: Sedan; 1: Truck; 2: Bus}; Vehicle usage nature: x 7 = {0: Operating; 1: Non-operating}; Number of times of annual inspection overdue: x 8 = Number of times of annual inspection overdue within a preset past time; Using the random sampling method, 80% of the tagged data set is selected as the training set, and 20% is selected as the test set. The ratio of the number of label types 0 and 1 in the training set and the test set is the same. Using the random forest method to train the training set, setting the initial value of the number of decision trees of the random forest model to 30, the initial value of the maximum depth parameter of the decision tree to 5, and the initial value of the minimum number of samples required for node splitting to 1. Using the test set to calculate the accuracy rate and F1 score indicators, confirming the model effect, and through the parameter tuning method of random search, continuously adjusting the parameters during several iterative training processes for data training; Train a random forest model. For the suspected fatigue driving data collected in real time, query the relevant information of the driver and the vehicle, call the random forest model for prediction, and output the probability of suspected fatigue driving, denoted as V feature , V feature is a probability value between 0 and 1, representing the probability that the driver is in a fatigue driving state.

7. The method for determining fatigue driving based on multi-dimensional data according to claim 1, wherein In step S105, determining physiological fatigue driving includes: using the driver's ID card information to query the following data during the period from 23:00 the previous day to 5:00 the next morning of the driver: A. Whether there is face capture data; B. Obtaining the affiliated mobile phone number, querying the mobile phone call data, and returning the last call time; C. Obtaining the affiliated mobile phone number, querying the last use time of the mobile phone APP; D. Querying the transportation means riding information; E. Querying the Internet access end time in Internet cafes; F. Querying the venue registration time; Introducing the nighttime activity index N for determining physiological fatigue, and the expression is: ， Among them, δ is the indicator function, ∑δ1 represents the total number of snapshot events in the statistical time period, ∑δ2 represents the total number of registrations in the statistical time period, and ∑δ3 represents the total number of calls and mobile phone APP usage in the statistical time period; Obtain the historical baseline value λ of the night activity index N from historical data, and determine the night activity index threshold N threshold , and use hypothesis testing to determine whether the driver is in a fatigue risk state. The null hypothesis H0: N ≤ N threshold indicates that the driver is in a normal rest state, and the alternative hypothesis H1: N > N threshold indicates that the driver has a fatigue risk; P(N≤N threshold ) is calculated by the cumulative distribution function of the Poisson distribution, and the expression is: ， where P(N ≤ N threshold ) is the probability of N ≤ N threshold , that is, the probability that the driver is in a normal rest state; k is the specific number of occurrences; e is the natural constant; k! is the factorial of k; The probability calculation expression of the driver being at risk of fatigue is: 。 8. The method for determining fatigue driving based on multi-dimensional data according to claim 7, characterized in that: The determination of physiological fatigue driving also includes step S106, which is determined by multi-factor weights. Let the physiological fatigue accumulation factor be F accumulation , and its calculation expression is: ， Among them, is the weight coefficient; if the continuous driving time T continuous exceeds two hours and F accumulation > 1, it is determined as physiological fatigue driving.

9. The method for determining fatigue driving based on multi-dimensional data according to any one of claims 1 to 8, characterized in that: It also includes a step of triggering an alarm. Records of physiological fatigue driving and continuous fatigue driving are displayed on the display interface. After review, notifications are made, text messages and / or telephone voice are triggered to remind the driver to take a rest.

10. A system for determining fatigue driving based on multi-dimensional data, characterized in that, include: Unified docking module for acquiring multi-dimensional perception data; Big data storage module, used to store accessed multi-dimensional sensing data; Big data analysis module, used to analyze the received perception data and judge physiological fatigue driving and continuous fatigue driving; The warning reminder module displays information on physiological fatigue driving and continuous fatigue driving and triggers alarms.

Images