Vehicle safety risk dynamic assessment and hierarchical early warning system and method based on multi-dimensional perception fusion

Abstract

The application relates to the technical field of intelligent traffic and Internet of Vehicles, in particular to a vehicle safety risk dynamic evaluation and hierarchical early warning system and method based on multi-dimensional perception fusion, which comprises a data acquisition module, a data fusion processing module, a risk evaluation calculation module and a hierarchical early warning and output module.The vehicle safety risk dynamic evaluation and hierarchical early warning system and method based on multi-dimensional perception fusion avoid single-index misjudgment or omission and realize multi-dimensional comprehensive evaluation by fusing positioning, active safety, emergency braking and environmental data; the system realizes the transformation from manual judgment alarm to automatic scoring / hierarchical disposal by a dynamic scoring system; monitoring personnel can directly focus on low-score vehicles, improving management efficiency; the parameters such as the weight of each dimension, the reference threshold and the alarm weight can be configured and adjusted, different scoring rules and threshold values can be configured according to different transport enterprises, and the system has strong adaptability.

Description

Technical Field

[0001] This invention relates to the fields of intelligent transportation and vehicle networking technology, and in particular to a system and method for dynamic assessment and graded early warning of vehicle safety risks based on multi-dimensional perception fusion. Background Technology

[0002] Currently, the assessment of vehicle safety risks mainly relies on alarms directly detected and reported by in-vehicle terminal equipment (such as Beidou positioning and vision-based active security equipment). For example, patent CN115909754A discloses a method for identifying motor vehicle violations based on Beidou high-precision positioning; patent CN119037415A mainly relies on visual data and bus data.

[0003] These solutions rely on a single data dimension and lack a comprehensive understanding of road conditions, vehicle status, and driver behavior. Furthermore, existing technologies cannot integrate and analyze vehicle status, driver behavior, and driving trajectory data, lacking dynamic cumulative assessment of risky behaviors and the ability to quantify long-term risk trends. In addition, existing technologies employ simplistic control strategies, lacking the ability to dynamically adjust vehicle control based on risk levels. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a vehicle safety risk dynamic assessment and graded early warning system based on multi-dimensional perception fusion. This invention also provides a vehicle safety risk dynamic assessment and graded early warning method based on multi-dimensional perception fusion, solving the problems of single safety evaluation dimensions and lack of intelligent graded alarm processing in existing vehicle monitoring platforms.

[0005] The technical solution adopted by the present invention to solve the above-mentioned technical problems is as follows: On the one hand, this invention provides a vehicle safety risk dynamic assessment and graded early warning system based on multi-dimensional perception fusion, comprising: Data acquisition module: The vehicle terminal uploads multi-source heterogeneous data to the cloud platform in real time to complete the acquisition of multi-source heterogeneous data; Data fusion processing module: used for cleaning and aligning multi-source heterogeneous data; Risk assessment calculation module: used to confirm the risk level through comprehensive risk assessment and event hazard level assessment; Tiered early warning and output module: Used to issue tiered early warnings based on risk levels and output corresponding data.

[0006] Furthermore, the tiered early warning and output module includes a real-time push unit, a daily / monthly report generation unit, and a blacklist management unit; Among them, the real-time push unit continuously receives alarm streams, calculates the event risk factor, and immediately triggers graded early warnings; Daily / Monthly Report Generation Unit: Generates daily reports daily and monthly reports monthly. The blacklist management unit updates the blacklist status.

[0007] On the other hand, the present invention also provides a method for dynamic assessment and graded early warning of vehicle safety risks based on multi-dimensional perception fusion, comprising the following steps: Step 1: Obtain multi-source heterogeneous data; Step 2: Process the acquired multi-source heterogeneous data; Step 3: Perform risk assessment calculations on the processed data; Step 4: Tiered early warning and response.

[0008] Furthermore, in step 3, the risk assessment includes a comprehensive risk assessment and an event hazard level assessment; Among them, the comprehensive risk assessment is achieved by weighting six real-time risk indices output by the six-dimensional risk index calculation model, and the event hazard level assessment is achieved by risk classification standards; The six risk indices are: fatigue index, distraction index, stability index, compliance index, road risk index, and accident index. The comprehensive risk assessment includes a daily comprehensive risk index and a monthly comprehensive risk index.

[0009] Furthermore, the daily comprehensive risk index is calculated as follows: Overall Risk Index = Fatigue Index * 25% + Distraction Index * 20% + Aggressiveness Index * 25% + Violation Index * 15% + Road Risk Index * 15%; The monthly comprehensive risk index is calculated as follows: The overall risk index is calculated as follows: Fatigue Index * 25% + Distraction Index * 20% + Aggressiveness Index * 20% + Violation Index * 15% + Road Risk Index * 10% + Accident Index * 10%.

[0010] Furthermore, the fatigue index = MIN((number of fatigue alarms per 100 km / baseline value × 100) × 40% + (number of severe fatigue alarms / baseline value × 100) × 20% + (cumulative overtime driving time / number of statistical days × 100) × 40%); Distraction Index = MIN(Number of distraction alarms per 100 km × 5, 100); Stability Index = (FCW score × 15%) + (HMW score × 5%) + (PCW score × 15%) + (Emergency Braking score × 20%) + (Rapid Acceleration score × 15%) + (Rapid Deceleration score × 15%) + (Sharp Turn score × 15%). Each sub-item score = MIN(number of times per 100 kilometers × corresponding coefficient, 100), with the following coefficients: FCW=5, HMW=2, PCW=5, emergency braking=5, rapid acceleration=5, rapid deceleration=5, sharp turn=5; Compliance Index = (Score for speeding duration × 20%) + (Score for nighttime driving restrictions × 20%) + (Score for smoking alarm × 20%) + (Score for making and receiving phone calls × 20%) + (Score for camera obstruction × 20%). Speeding duration percentage score = MIN(speeding duration / driving time / 5% × 100, 100); Night driving restriction frequency score = MIN(night driving restriction alarm frequency / number of statistical days × 20, 100); Smoking / phone call / obstruction score = MIN(number per 100 kilometers × 5, 100); Road risk index = (score for highway mileage percentage × 40%) + (score for nighttime driving time percentage × 40%) + (score for severe weather percentage × 20%). Highway mileage percentage score = MIN(highway mileage / total mileage / 30% × 100, 100); Night driving time score = MIN(night driving time / total driving time / 20% × 100, 100); Severe weather percentage score = MIN(severe weather driving time / total driving time / 10% × 100, 100); Accident Index = (Score of at-fault accidents * 50%) + (Score of accident severity * 30%) + (Score of cumulative number of historical accidents * 20%) Score for at-fault accidents = MIN (Number of at-fault accidents * 40, 100); Accident severity score = MIN(Number of personal injury accidents / Number of accidents * 100 + Number of property damage accidents / Number of accidents * 50, 100); Historical accident cumulative score = MIN (total number of at-fault accidents in the past three years * 20, 100) Furthermore, the comprehensive risk score = fatigue index × 25% + distraction index × 20% + stability index × 25% + compliance index × 15% + road risk index × 15%; The comprehensive risk classification standard is as follows: [0, 30]: Low risk; (30, 60): Medium risk; (60, 80): High risk; (80, 100): Extremely high risk; Furthermore, event hazard level assessment: For each alarm event, calculate the event hazard coefficient, and confirm the hazard level based on the hazard coefficient; Event risk factor = alarm weight × frequency factor × combination factor × environmental factor × vehicle speed factor; Warning weights: Emergency braking = 10, Fatigue = 8, Distraction = 6, Forward collision = 5, Pedestrian collision = 5, Other = 2; Frequency coefficient: 3 or more alarms of the same type within 5 minutes × 1.5, 5 or more × 2; Combination coefficient: ≥2 alarm types within 1 minute × 1.5, ≥3 types × 2; Environmental factors: Nighttime ×1.5, Highway ×1.2, Severe weather ×2; Speed ​​coefficient: 1.5 for speeds > 80 km / h, 2 for speeds > 100 km / h; The risk levels are categorized as extremely dangerous, high dangerous, medium dangerous, and low dangerous. The criteria for classifying hazard levels are as follows: ≥60 is extremely risky, ≥30 is high risky, ≥15 is medium risky, and <15 is low risky.

[0011] Furthermore, in step 4, a graded early warning system is implemented based on the level of danger; The incident is classified as extremely dangerous: mark it as a priority; notify the driver and fleet manager; and mark the incident as requiring manual review. The incident is classified as high-risk: it has been marked as a priority case and a pop-up notification has been sent. The incident is classified as medium-risk: mark the vehicle and request intervention to investigate; The incident is classified as low-risk: it is only recorded in reports and does not interfere with normal driving.

[0012] Furthermore, in step 1, multi-source heterogeneous data is acquired in real time through the vehicle terminal. The multi-source heterogeneous data includes: ADAS alarm data, DMS alarm data, AEBS trigger information, positioning data, and CAN data.

[0013] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention relates to a vehicle safety risk dynamic assessment and graded early warning system and method based on multi-dimensional perception fusion. By integrating positioning, active safety, emergency braking, and environmental data, it avoids misjudgment or omission of single indicators and achieves comprehensive multi-dimensional assessment. Through a dynamic scoring system, it supports daily and monthly rolling calculations, enabling timely detection of rising risk trends and helping managers intervene early. It represents a shift from manual alarm judgment to automatic scoring / graded handling by the system. Monitoring personnel no longer need to handle a large number of chaotic alarms but can directly focus on low-scoring vehicles, improving management efficiency. Parameters such as the weight of each dimension, baseline threshold, and alarm weight are configurable and adjustable, allowing for different scoring rules and thresholds to be configured for different transportation companies, demonstrating strong adaptability. Attached Figure Description

[0014] Figure 1This is a flowchart of the method for dynamic assessment and graded early warning of vehicle safety risks based on multi-dimensional perception fusion, as described in this invention. Detailed Implementation

[0015] The technical solution of the present invention will be further described in a non-limiting manner below with reference to the accompanying drawings and specific embodiments.

[0016] Example 1: A vehicle safety risk dynamic assessment and graded early warning system based on multi-dimensional perception fusion, comprising: Data acquisition module: The vehicle terminal uploads multi-source heterogeneous data to the cloud platform in real time, completing the acquisition of multi-source heterogeneous data.

[0017] Data fusion and processing module: Used for cleaning and aligning multi-source heterogeneous data. The cloud platform performs data cleaning on a regular basis and stores the data in the data warehouse.

[0018] Risk assessment calculation module: Used to confirm risk levels through comprehensive risk assessment and event hazard level assessment. It triggers batch calculation tasks daily to update the five-dimensional index and comprehensive risk score for all drivers.

[0019] Tiered early warning and output module: Used to issue tiered early warnings based on risk levels and output corresponding data.

[0020] The tiered early warning and output module includes a real-time push unit, a daily / monthly report generation unit, and a blacklist management unit.

[0021] Among them, the real-time event processing module pushes alarms continuously, calculates the event risk factor, and immediately triggers graded warnings.

[0022] Daily / Monthly Report Generation Unit: Generates daily reports daily and monthly reports monthly. The blacklist management unit updates the blacklist status.

[0023] Prerequisites: This system requires the vehicle to be equipped with an active safety control system terminal with Beidou positioning function and an emergency braking system in advance. The terminal equipment can connect to the network and upload speed, location, ADAS alarm, DSM alarm and AEBS emergency braking data.

[0024] Example 2: like Figure 1 As shown, a method for dynamic assessment and graded early warning of vehicle safety risks based on multi-dimensional perception fusion includes the following steps: Step 1: Obtain multi-source heterogeneous data.

[0025] Multi-source heterogeneous data is acquired in real time via an in-vehicle terminal. This multi-source heterogeneous data includes: ADAS warning data (forward collision warning, lane departure warning, close following distance, pedestrian collision, etc.); DMS alarm data (smoking, fatigue, distraction, camera obstruction); AEBS trigger information (trigger time, vehicle speed, relative speed, deceleration, braking result); Location data (speed, distance, latitude and longitude, altitude); CAN data (number of times of rapid acceleration, rapid deceleration, and sharp turns).

[0026] Step 2: Process the acquired multi-source heterogeneous data.

[0027] The process is as follows: First, the acquired heterogeneous data from multiple sources is cleaned and time-aligned: Data cleaning includes filtering outlier data, handling null values, and performing logical checks.

[0028] Filter out abnormal data, including: missing time, location time < current time, speed > 160 km / h, location drift, and repeated reporting of the same alarm within 3 seconds.

[0029] Null value handling includes: for missing velocity values, interpolation is used to fill in the missing values ​​based on the preceding and following points; for missing latitude and longitude values, the missing values ​​are filled in using the preceding and following trajectories. Logical verification is performed, and suspicious cases are marked for manual review. This includes: alarm type and vehicle speed discrepancies, and braking triggering without corresponding warnings.

[0030] Time alignment refers to sorting all data by time, based on the location time.

[0031] Then, remove false positive data: Remove alarm records marked as false alarms and summarize mileage and driving time on a daily basis.

[0032] Finally, the data after removing false alarms is input into the thematic data model, which then performs data classification and summarization to achieve data fusion.

[0033] Data is categorized and summarized on a daily basis, including the number of warnings per 100 kilometers for fatigue driving, distracted driving, forward collisions, vehicle deviation, following too closely, emergency braking, rapid acceleration / deceleration / sharp turns, smoking / talking on the phone / obstructing cameras, speeding duration, number of times prohibited from driving at night, highway mileage, nighttime driving time, and driving time in adverse weather conditions. Specifically, the subject data model is the result obtained after secondary processing such as classifying and summarizing the collected source data according to the business subject. For example, the fatigue index formula uses the number of fatigue alarms per 100 kilometers, which is the processed data. The source data is fatigue alarms and mileage. The number of fatigue alarms per 100 kilometers = number of fatigue alarms / mileage * 100.

[0034] Step 3: Perform risk assessment calculations on the data processed in Step 2.

[0035] Risk assessment includes comprehensive risk assessment and event hazard level assessment.

[0036] Among them, the comprehensive risk assessment is achieved by weighting six real-time risk indices output by the six-dimensional risk index calculation model, and the event hazard level assessment is achieved by risk classification standards.

[0037] The six-dimensional risk index calculation model includes six risk indices, each with a score range of 0-100. The higher the score of an index, the higher the risk.

[0038] The six risk indices are: fatigue index, distraction index, stability index, compliance index, road risk index, and accident index.

[0039] The comprehensive risk assessment includes a daily comprehensive risk index and a monthly comprehensive risk index.

[0040] The daily comprehensive risk index is calculated as follows: Overall Risk Index = Fatigue Index * 25% + Distraction Index * 20% + Aggressiveness Index * 25% + Violation Index * 15% + Road Risk Index * 15% The daily comprehensive risk index does not include the accident index, as accidents are sporadic and can significantly impact the results.

[0041] The monthly comprehensive risk index is calculated as follows: Overall Risk Index = Fatigue Index * 25% + Distraction Index * 20% + Aggressiveness Index * 20% + Violation Index * 15% + Road Risk Index * 10% + Accident Index * 10% Fatigue Index = MIN((Number of fatigue alarms per 100 km / Baseline value × 100) × 40% + (Number of severe fatigue alarms / Baseline value × 100) × 20% + (Cumulative overtime driving time / Number of days in the statistics × 100) × 40%).

[0042] Distraction Index = MIN (Number of distraction alarms per 100 km × 5, 100).

[0043] Stability Index = (FCW score × 15%) + (HMW score × 5%) + (PCW score × 15%) + (Emergency Braking score × 20%) + (Rapid Acceleration score × 15%) + (Rapid Deceleration score × 15%) + (Sharp Turn score × 15%).

[0044] The score for each item is calculated as MIN(number of times per 100 kilometers × corresponding coefficient, 100). The coefficients are as follows: FCW=5, HMW=2, PCW=5, emergency braking=5, rapid acceleration=5, rapid deceleration=5, and sharp turn=5.

[0045] Compliance Index = (Score for speeding duration × 20%) + (Score for nighttime driving restrictions × 20%) + (Score for smoking alarm × 20%) + (Score for making and receiving phone calls × 20%) + (Score for camera obstruction × 20%).

[0046] Speeding duration percentage score = MIN(speeding duration / driving time / 5% × 100, 100); Nighttime driving restriction frequency score = MIN(nighttime driving restriction alarm frequency / number of statistical days × 20, 100); Smoking / phone call / obstruction score = MIN(number per 100 kilometers × 5, 100).

[0047] Road risk index = (score for highway mileage percentage × 40%) + (score for nighttime driving time percentage × 40%) + (score for severe weather percentage × 20%).

[0048] Highway mileage percentage score = MIN(highway mileage / total mileage / 30% × 100, 100); Night driving time score = MIN(night driving time / total driving time / 20% × 100, 100); Severe weather percentage score = MIN(severe weather driving time / total driving time / 10% × 100, 100).

[0049] Accident Index = (Score of at-fault accidents * 50%) + (Score of accident severity * 30%) + (Score of cumulative number of historical accidents * 20%) Score for at-fault accidents = MIN (Number of at-fault accidents * 40, 100); Accident severity score = MIN(Number of personal injury accidents / Number of accidents * 100 + Number of property damage accidents / Number of accidents * 50, 100); Historical accident cumulative score = MIN (total number of at-fault accidents in the past three years * 20, 100) The aforementioned dimensional weights, benchmark thresholds, coefficients, and other parameters can be adjusted through the configuration file.

[0050] Overall risk score = Fatigue index × 25% + Distraction index × 20% + Stability index × 25% + Compliance index × 15% + Road risk index × 15%.

[0051] The comprehensive risk classification standard is as follows: [0, 30]: Low risk; (30, 60): Medium risk; (60, 80): High risk; (80, 100): Extremely high risk; Incident hazard level assessment: For each alarm event (including ADAS, DMS, AEBS), calculate the event risk coefficient and determine the risk level based on the risk coefficient.

[0052] Event risk factor = Alarm weight × Frequency factor × Combination factor × Environmental factor × Vehicle speed factor Warning weights: Emergency braking = 10, Fatigue = 8, Distraction = 6, Forward collision = 5, Pedestrian collision = 5, Other = 2.

[0053] Frequency coefficient: 1.5 for more than 3 alarms of the same type within 5 minutes, and 2 for more than 5 alarms.

[0054] Combination coefficient: ≥2 alarm types within 1 minute × 1.5, ≥3 types × 2.

[0055] Environmental factor: Nighttime ×1.5, Highway ×1.2, Severe weather ×2.

[0056] Speed ​​coefficient: 1.5 for speeds > 80 km / h, 2 for speeds > 100 km / h.

[0057] The risk levels are categorized as extremely dangerous, high dangerous, medium dangerous, and low dangerous. The criteria for classifying hazard levels are as follows: ≥60 is extremely risky, ≥30 is high risky, ≥15 is medium risky, and <15 is low risky.

[0058] Step 4: Tiered early warning and response.

[0059] Based on the level of danger, a graded early warning system will be implemented.

[0060] The incident is classified as extremely dangerous: a purple alert is generated; it is displayed prominently on the monitoring screen and marked as a priority; the driver and fleet manager are notified via in-vehicle voice, SMS, and other means; and the incident is marked as requiring manual review.

[0061] The incident is classified as high-risk: a red alert is generated; it is displayed prominently on the monitoring screen, marked as a key area of ​​concern, and a pop-up notification is sent to the enterprise security administrator.

[0062] The incident is classified as medium risk: an orange alert has been generated. The vehicle has been flagged, and monitoring personnel have been alerted to investigate.

[0063] The incident is classified as low-risk: a yellow alert is generated. It is only recorded in the report and does not interfere with normal driving.

[0064] Report push: Daily / monthly reports are generated, listing high-risk and extremely dangerous events, as well as the top 10% of drivers based on their comprehensive risk scores.

[0065] Blacklist mechanism: Drivers with a comprehensive risk score >60 for two consecutive months and who have been involved in extremely dangerous incidents will be placed on the observation list; drivers with a comprehensive risk score >60 for three consecutive months and who are at fault in accidents will be placed on the high-risk list.

[0066] Data output: Provides a standardized API to output the driver's five-dimensional index, comprehensive risk score, and blacklist level.

[0067] The above description is an explanation of the present invention and not a limitation thereof. The scope of the present invention is defined by the claims. Within the scope of protection of the present invention, any form of modification may be made.

Claims

1. A vehicle safety risk dynamic assessment and hierarchical warning system based on multi-dimensional perception fusion, characterized in that, include: Data acquisition module: The vehicle terminal uploads multi-source heterogeneous data to the cloud platform in real time to complete the acquisition of multi-source heterogeneous data; Data fusion processing module: used for cleaning and aligning multi-source heterogeneous data; Risk assessment calculation module: used to confirm the risk level through comprehensive risk assessment and event hazard level assessment; Tiered early warning and output module: Used to issue tiered early warnings based on risk levels and output corresponding data.

2. The vehicle safety risk dynamic assessment and graded early warning system according to claim 1, characterized in that, The tiered early warning and output module includes a real-time push unit, a daily / monthly report generation unit, and a blacklist management unit; Among them, the real-time push unit continuously receives alarm streams, calculates the event risk factor, and immediately triggers graded early warnings; Daily / Monthly Report Generation Unit: Generates daily reports daily and monthly reports monthly. The blacklist management unit updates the blacklist status.

3. A method for dynamic assessment and graded early warning of vehicle safety risks based on multi-dimensional perception fusion, characterized in that, The system applied to the vehicle safety risk dynamic assessment and graded early warning system as described in claim 1 or 2 includes the following steps: Step 1: Obtain multi-source heterogeneous data; Step 2: Process the acquired multi-source heterogeneous data; Step 3: Perform risk assessment calculations on the processed data; Step 4: Tiered early warning and response.

4. The method for dynamic assessment and graded early warning of vehicle safety risks according to claim 3, characterized in that, In step 3, the risk assessment includes a comprehensive risk assessment and an event hazard level assessment; Among them, the comprehensive risk assessment is achieved by weighting six real-time risk indices output by the six-dimensional risk index calculation model, and the event hazard level assessment is achieved by risk classification standards; The six risk indices are: fatigue index, distraction index, stability index, compliance index, road risk index, and accident index. The comprehensive risk assessment includes a daily comprehensive risk index and a monthly comprehensive risk index.

5. The method for dynamic assessment and graded early warning of vehicle safety risks according to claim 4, characterized in that, The daily comprehensive risk index is calculated as follows: Overall Risk Index = Fatigue Index * 25% + Distraction Index * 20% + Aggressiveness Index * 25% + Violation Index * 15% + Road Risk Index * 15%; The monthly comprehensive risk index is calculated as follows: The overall risk index is calculated as follows: Fatigue Index * 25% + Distraction Index * 20% + Aggressive Index * 20% + Violation Index * 15% + Road Risk Index * 10% + Accident Index * 10%.

6. The method for dynamic assessment and graded early warning of vehicle safety risks according to claim 5, characterized in that, Fatigue Index = MIN((Number of fatigue alarms per 100 km / Baseline value × 100) × 40% + (Number of severe fatigue alarms / Baseline value × 100) × 20% + (Cumulative overtime driving time / Number of days in the statistics × 100) × 40%); Distraction Index = MIN(Number of distraction alarms per 100 km × 5, 100); Stability Index = (FCW score × 15%) + (HMW score × 5%) + (PCW score × 15%) + (Emergency Braking score × 20%) + (Rapid Acceleration score × 15%) + (Rapid Deceleration score × 15%) + (Sharp Turn score × 15%). Each sub-item score = MIN(number of times per 100 kilometers × corresponding coefficient, 100), with the following coefficients: FCW=5, HMW=2, PCW=5, emergency braking=5, rapid acceleration=5, rapid deceleration=5, sharp turn=5; Compliance Index = (Score for speeding duration × 20%) + (Score for nighttime driving restrictions × 20%) + (Score for smoking alarm × 20%) + (Score for making and receiving phone calls × 20%) + (Score for camera obstruction × 20%). Speeding duration percentage score = MIN(speeding duration / driving time / 5% × 100, 100); Night driving restriction frequency score = MIN(night driving restriction alarm frequency / number of statistical days × 20, 100); Smoking / phone call / obstruction score = MIN(number per 100 kilometers × 5, 100); Road risk index = (score for highway mileage percentage × 40%) + (score for nighttime driving time percentage × 40%) + (score for severe weather percentage × 20%). Highway mileage percentage score = MIN(highway mileage / total mileage / 30% × 100, 100); Night driving time score = MIN(night driving time / total driving time / 20% × 100, 100); Severe weather percentage score = MIN(severe weather driving time / total driving time / 10% × 100, 100); Accident Index = (Score of at-fault accidents * 50%) + (Score of accident severity * 30%) + (Score of cumulative number of historical accidents * 20%) Score for at-fault accidents = MIN (Number of at-fault accidents * 40, 100); Accident severity score = MIN(Number of personal injury accidents / Number of accidents * 100 + Number of property damage accidents / Number of accidents * 50, 100); Historical accident cumulative score = MIN (total number of at-fault accidents in the past three years * 20, 100).

7. The method for dynamic assessment and graded early warning of vehicle safety risks according to claim 6, characterized in that, Overall risk score = Fatigue index × 25% + Distraction index × 20% + Stability index × 25% + Compliance index × 15% + Road risk index × 15%; The comprehensive risk classification standard is as follows: [0, 30]: Low risk; (30, 60): Medium risk; (60, 80): High risk; (80, 100): Extremely high risk.

8. The method for dynamic assessment and graded early warning of vehicle safety risks according to claim 7, characterized in that, Event risk level assessment: For each alarm event, calculate the event risk coefficient and determine the risk level based on the risk coefficient; Event risk factor = alarm weight × frequency factor × combination factor × environmental factor × vehicle speed factor; Warning weights: Emergency braking = 10, Fatigue = 8, Distraction = 6, Forward collision = 5, Pedestrian collision = 5, Other = 2; Frequency coefficient: 3 or more alarms of the same type within 5 minutes × 1.5, 5 or more × 2; Combination coefficient: ≥2 alarm types within 1 minute × 1.5, ≥3 types × 2; Environmental factors: Nighttime ×1.5, Highway ×1.2, Severe weather ×2; Speed ​​coefficient: 1.5 for speeds > 80 km / h, 2 for speeds > 100 km / h; The risk levels are categorized as extremely dangerous, high dangerous, medium dangerous, and low dangerous. The criteria for classifying hazard levels are as follows: ≥60 is extremely risky, ≥30 is high risky, ≥15 is medium risky, and <15 is low risky.

9. The method for dynamic assessment and graded early warning of vehicle safety risks according to claim 3, characterized in that, In step 4, a graded warning is issued based on the level of danger; The incident is classified as extremely dangerous: mark it as a priority; notify the driver and fleet manager; and mark the incident as requiring manual review. The incident is classified as high-risk: it has been marked as a priority case and a pop-up notification has been sent. The incident is classified as medium-risk: mark the vehicle and request intervention to investigate; The incident is classified as low-risk: it is only recorded in reports and does not interfere with normal driving.

10. The method for dynamic assessment and graded early warning of vehicle safety risks according to claim 3, characterized in that, In step 1, multi-source heterogeneous data is acquired in real time through the vehicle terminal. The multi-source heterogeneous data includes: ADAS alarm data, DMS alarm data, AEBS trigger information, positioning data, and CAN data.

Images