Bayesian-based method for assessing road and vehicle-road matching risk index in autonomous driving

Abstract

This invention presents a Bayesian-based method for assessing the risk index of autonomous driving roads and vehicle-road matching, addressing the technical challenge of effectively evaluating the risk level of autonomous driving roads and vehicle-road matching. The method involves processing and analyzing autonomous driving data using computers and supporting software, including data cleaning, constructing a Bayesian network model, slicing roads, and segmenting the data to assess road risk points, road nodes, and the risk level and vehicle-road matching risk index. The advantages of this invention are that it effectively supports autonomous driving road planning, enabling risk assessment of the route in the early stages of route selection, eliminating high-risk road sections, and improving road safety.

Description

Technical Field

[0001] This invention belongs to the field of autonomous driving risk assessment technology, and relates to assessment methods for autonomous driving road risks and autonomous driving vehicle-road matching risks, particularly a Bayesian-based method for assessing autonomous driving road and vehicle-road matching risks. Background Technology

[0002] Autonomous vehicles have transitioned from testing and demonstration to commercial application. Analysis reveals that the probability of accidents involving autonomous vehicles is increasing year by year, both domestically and internationally, and several application issues urgently need to be addressed. Firstly, there is a lack of reasonable technical means for setting up autonomous driving routes / areas; secondly, the matching technology between the design operating conditions and operating routes / areas of autonomous vehicles is still under exploration. Risk assessment technologies for autonomous driving roads and vehicle-to-infrastructure (V2I) matching need to be implemented and improved. Summary of the Invention

[0003] To address the technical problem of effectively assessing autonomous driving road risks and vehicle-road matching risks, this invention designs a Bayesian-based method for evaluating autonomous driving road and vehicle-road matching risk indices. This method comprehensively analyzes and evaluates the risk situation through Bayesian network models, data-driven approaches, and road slicing.

[0004] The technical solution adopted in this invention is a Bayesian-based method for assessing the road risk of autonomous driving. This method involves processing and analyzing autonomous driving data using a computer and supporting software. Crucially, the specific steps of this method are as follows:

[0005] a1. Process the autonomous driving data to form a database that includes accident injury level, road type, road surface dryness and wetness, road speed limit, road condition, accident type, weather and lighting elements. Among them, the accident injury level element includes four items: no injury, minor injury, moderate injury and severe injury.

[0006] b1. Based on the elements and content in the database in step a1, take the accident injury severity element as the target node and all other elements as nodes, and construct a Bayesian network risk qualitative assessment model in the supporting software program.

[0007] c1. The Bayesian network risk qualitative assessment model is used to learn and train all accident data in the database to obtain the joint distribution of conditional probabilities of all nodes, i.e., each element, thus constructing the Bayesian network risk quantitative assessment model.

[0008] d1. The autonomous driving roads to be evaluated are sliced ​​into multiple continuous road slices. The road slices are split in the following two ways: the same road type is split according to the unit mileage; different road types are split according to the differences in road characteristics.

[0009] e1. Extract the road element set for each road slice according to the elements in the Bayesian network risk quantitative assessment model;

[0010] f1. Based on the road element set of each road slice, conduct risk assessments on each road slice to form road risk points. Specifically, using the elements of the road slice as known factors, and according to the joint distribution of conditional probabilities in the Bayesian network risk quantitative assessment model constructed in step c1, calculate the probability of no injury, minor injury, moderate injury, and severe injury in the accident injury severity of the target node according to Bayesian network theory, and quantify the risk of the road slice. The risk quantification formula for any road slice is as follows:

[0011] r i =r NoIi ×w NoI +r Mini ×w Min +r Modi ×w Mod +r Seri ×w Ser Formula 1

[0012] In Equation 1, r i Risk assessment results for the i-th road slice.

[0013] r NoIi For the i-th road slice, the probability of no damage is...

[0014] r Mini Regarding the probability that the i-th road slice is slightly damaged,

[0015] r Modi : The probability of moderate damage for the i-th road slice

[0016] r Seri : The probability of severe damage to the i-th road slice

[0017] w NoI w Min w Mod w Ser : These correspond to the risk weights of the four categories of injury severity in the target node accident: no injury, minor injury, moderate injury, and severe injury, respectively, with values ​​increasing sequentially. If r i>If the risk threshold constant for a road slice is constant, then the i-th road slice is considered a risky road slice, and the set of all risky slices is the road risk point;

[0018] g1. Calculate the overall road risk of the autonomous driving road based on the risk level of any road slice obtained in step f1. The overall road risk quantification formula is as follows:

[0019]

[0020] In Equation 2, R: the quantitative index of the road risk being assessed.

[0021] n: The number of road segments in this road.

[0022] x i : The length of the i-th road slice.

[0023] l: The total length of the road.

[0024] r Max Risk value of the road segment with the highest risk in this road.

[0025] In step e1, the element extraction process uses equipment equipped with a visual sensor to collect data, or it is collected by querying design and construction materials in conjunction with the site.

[0026] In step a1, the data cleaning process includes deleting incomplete data, deleting duplicate data, deleting interfering data, deleting irrelevant data, and performing data mining to construct indirect elements.

[0027] This invention also relates to a Bayesian-based method for assessing the risk index of autonomous vehicle-road matching. The method is implemented by processing and analyzing autonomous driving data using a computer and supporting software programs. Crucially, the specific steps of the method are as follows:

[0028] a2. Process the autonomous driving data to form a database that includes accident injury level, road type, road surface dryness and wetness, road speed limit, road condition, accident type, weather and lighting elements. Among them, the accident injury level element includes four items: no injury, minor injury, moderate injury and severe injury.

[0029] b2. Based on the elements and content in the database in step a2, take the accident injury severity element as the target node and all other elements as nodes, and construct a Bayesian network risk qualitative assessment model in the supporting software program.

[0030] c2. The Bayesian network risk qualitative assessment model is used to learn and train all accident data in the database to obtain the joint distribution of conditional probabilities of all nodes, i.e., each element, thus constructing the Bayesian network risk quantitative assessment model.

[0031] d2. The autonomous driving roads to be evaluated are sliced ​​into multiple continuous road slices. The road slices are split in the following two ways: the same road type is split according to the unit mileage; different road types are split according to the differences in road characteristics.

[0032] e2. Extract the road element set for each road slice according to the elements in the Bayesian network risk quantitative assessment model;

[0033] f2. Based on the Bayesian network risk quantitative assessment model in step c2, according to the Bayesian network technology theory, the sensitivity of all road element nodes and target nodes to the degree of accident damage in the Bayesian model is obtained. After normalizing the sensitivity, the risk weight of all elements to the degree of accident damage is obtained.

[0034] g2. Calculate the vehicle-road segment matching risk index between each road slice and the vehicle-road segment to be operated by the autonomous vehicle. The calculation formula is as follows:

[0035]

[0036] In Equation 3, m v The risk index for matching vehicle-road slices in the v-th road slice.

[0037] q: The number of features in the v-th road slice.

[0038] e j In the v-th road slice, the j-th element is assigned a value of 1 if it matches the design operating conditions of the autonomous vehicle, and 0 if it does not match the design operating conditions of the autonomous vehicle.

[0039] w j The risk weight of the j-th element in the v-th road slice obtained in step f2.

[0040] When any road slice's vehicle-road slice matches the risk index m v When the value is less than the vehicle-road segment matching risk threshold constant, the segment of the road is determined to be the vehicle-road matching risk point of the vehicle. The set of all vehicle-road matching risk points in the road is the vehicle-road matching risk point.

[0041] h2. Calculate the vehicle-road matching risk index between the proposed operating road and the proposed autonomous vehicles. The calculation formula is as follows:

[0042]

[0043] In Equation 4, M: the vehicle-road matching risk index of the road.

[0044] p: The number of road segments in this road.

[0045] q: The number of elements in the road slice of this road.

[0046] x k : The length of the k-th road slice.

[0047] l: The total length of the road.

[0048] m kj In this road, for the k-th road slice, the j-th element's matching value is 1 if it matches the design operating conditions of the autonomous vehicle, and 0 if it does not match the design operating conditions of the autonomous vehicle.

[0049] w kj The weight of the j-th element in the k-th road slice of this road.

[0050] When the vehicle-road matching risk index M between the proposed operating road and the proposed autonomous vehicle is less than the vehicle-road matching risk threshold constant, it is determined that there is a risk in the autonomous vehicle operating on that road.

[0051] In step e2, the element extraction process uses equipment equipped with a visual sensor to collect data, or it is collected by querying design and construction materials in conjunction with the site.

[0052] In step a2, the data cleaning process includes deleting incomplete data, deleting duplicate data, deleting interfering data, deleting irrelevant data, and performing data mining to construct indirect elements.

[0053] The core technological innovation of this invention lies in its continuous learning and training on a constantly updated autonomous driving accident database based on Bayesian network theory. This yields a probabilistic joint distribution of various road feature elements and accident severity that approximates reality. Bayesian causal inference is then used to assess and predict the risks of other roads. Furthermore, based on the probabilistic joint distribution and Bayesian network sensitivity analysis, risk weights for different elements are derived. The matching of each element in each road slice with the design operating conditions of the autonomous vehicle is weighted, ultimately calculating the vehicle-road matching risk index.

[0054] The technical advantages of this invention lie in its ability to slice roads and accurately identify road risk points, effectively supporting the planning of roads for autonomous driving. In the early stages of route selection, risk assessments are conducted on the operating route, eliminating high-risk sections and improving road safety. A road risk quantification index reflects the risk status of different operating roads, effectively supporting risk classification. The vehicle-road matching risk index reflects the matching risk between autonomous vehicles and the intended autonomous driving road; the risk level of autonomous vehicle operation can be controlled by adjusting the threshold of the vehicle-road matching risk index. The vehicle-road slice matching risk index reflects the risk level between autonomous vehicles and a specific road slice within the intended autonomous driving road; by modifying or rerouting road slices with abnormal vehicle-road slice matching risk indices, the overall risk level between autonomous vehicles and the intended autonomous driving road can be reduced, thus lowering the overall risk of autonomous vehicle-road operation. Attached Figure Description

[0055] Figure 1 This is a flowchart of the risk assessment method of the present invention.

[0056] Figure 2 This is a schematic diagram of a specific implementation of constructing a Bayesian network model.

[0057] Figure 3 This is a schematic diagram of a specific implementation of a Bayesian network risk assessment model formed after training.

[0058] Figure 4 This is a schematic diagram illustrating a specific implementation of the risk assessment results for road slabs. Detailed Implementation

[0059] A Bayesian network is a directed acyclic graph network topology where each node represents a random variable, and the strength of the association between nodes is represented by a conditional probability table. The data-driven approach supports the iterative development of the Bayesian network qualitative analysis model through a continuously expanding database of autonomous driving accidents. The road slicing allows for refined risk assessment while flexibly controlling the granularity based on actual road conditions; for example, highways can use larger-granularity road slices, while urban roads require smaller-granularity slices, enabling the method to be applied to different types of roads.

[0060] In its specific implementation, this invention uses Incident-Reports-ADS (Autonomous Driving Data) from the National Highway Traffic Safety Administration (NHTSA) as an example. Combined with... Figures 1 to 4 The overall method and process of the present invention will be described in detail.

[0061] Step 1: First, perform data cleaning and mining on the Incident-Reports-ADS autonomous driving data to construct the NHTSA database. The data cleaning and mining process includes:

[0062] Delete incomplete data, and delete data whose key information is "UNKNOWN".

[0063] Remove duplicate data. For data with the same "Report ID", only keep the data with the highest Report Version. Remove duplicate data for the same incident.

[0064] Remove interfering data: remove data marked "NO" in "Automation System Engaged?"; remove data in "Narrative" indicating manual driving at the time of the accident; remove data marked "stopped" and "parked" in "SV PreCrash Movement".

[0065] Remove irrelevant elements, including those related to the investigation, such as the judge.

[0066] Data mining was conducted to construct indirect elements, and accident form elements, "Accident Form," were constructed based on "SV Contact Area," "CP Contact Area," and "Narrative" according to the accident investigation method.

[0067] Construct the NHTSA database, and the database elements are as follows:

[0068] Table 1. NHTSA Database Elements (Database Data Items)

[0069]

[0070]

[0071] Step 2: Conduct risk analysis and construct a Bayesian network risk qualitative assessment model.

[0072] Based on the accident data element categories and content in the NHTSA database, a Bayesian network model is constructed according to professional technical risk analysis. The specific model is as follows: Figure 2 The Bayesian network risk qualitative assessment model is as follows:

[0073] Element category = node

[0074] The Injury Severity node is the target node and the target of the model's computational output.

[0075] The nodes Road way, Posted Speed ​​Limit, Road way Description, Crash With, SVPreCrash Movement, CP PreCrash Movement, Roadway Surface, and Lighting serve as the parent nodes of the Accident Form node.

[0076] The Lighting node is the same parent node as the Accident Form node and the target node Injury Severity;

[0077] The Weather node is the same parent node as the Roadway Surface node and the target node Injury Severity;

[0078] The nodes Weather, Accident Form, Lighting, SVPrecrashSpeed, and Operator are the parent nodes of the target node Injury Severity.

[0079] Step 3: Based on the Bayesian network qualitative risk assessment model, conduct data learning from the NHTSA database to form a Bayesian network quantitative risk assessment model. This involves the Bayesian network model learning from all incident data in the NHTSA database. The accompanying software management program uses NETICA software to convert the NHTSA database to "CSV" format. The Bayesian network qualitative risk assessment model is then built within NETICA software, and the "Cases-Learn-Incorp Cae File" function is used to train the model on the NHTSA database. After training, a Bayesian network quantitative risk assessment model is formed, as shown below. Figure 3 As shown.

[0080] As the NHTSA's autonomous driving data, "Incident-Reports-ADS," continues to grow, the NHTSA database will also continue to expand. Therefore, the Bayesian network risk quantitative assessment model will continue to learn and become more accurate.

[0081] Step 4: Construct continuous road slices. There are two ways to construct continuous road slices:

[0082] Method 1: For roads of the same type or with similar road environments, the road is divided into slices based on unit mileage. The slices are continuous and do not have any intervals. The unit mileage can be arbitrarily set according to the needs, such as 100m, 500m, 1km, etc. This road slice is called a unit mileage road slice.

[0083] Method 2 addresses routes with multiple road types or significantly different road environments. Based on the differences in road characteristics, the routes are divided into continuous slices of varying lengths. These road slices are called length-weighted road slices.

[0084] Step 5: Extract the feature set of the road slices. Based on the features in the Bayesian network risk qualitative assessment model from Step 2, extract the road feature set of the road to be put into operation. Extraction methods can include using equipment equipped with visual sensors, or collecting data by querying design and construction materials in conjunction with on-site data collection.

[0085] Step 6: Conduct risk assessments on each slice individually and generate "Output 1: Road Risk Points".

[0086] Using the prior of a Bayesian network-based quantitative risk assessment model as the prior probability, and inputting all elements of each slice as conditional probability data into the model, the risk probability can be directly calculated. Further utilizing NETICA software, input can be directly applied to the Bayesian network-based quantitative risk assessment model, and by adjusting the probabilities of the determined elements to 100%, the risk probability can be directly output. For example, after inputting road slice elements such as "Street, Dry, Clear, DarkLighted," the model will adjust "Street, Dry, Clear, DarkLighted" to 100%, and the target "Injury Severity" will be output as the risk assessment result. Figure 4 As shown. The risk quantification formula for any road slice is:

[0087] r i =r NoIi ×w NoI +r Mini ×w Min +r Modi ×w Mod +r Seri ×w Ser Formula 1

[0088] In Equation 1, r i Risk assessment results for the i-th road slice.

[0089] r NoIi For the i-th road slice, the probability of no damage is...

[0090] r Mini Regarding the probability that the i-th road slice is slightly damaged,

[0091] r Modi : The probability of moderate damage for the i-th road slice

[0092] r Seri: The probability of severe damage to the i-th road slice

[0093] w NoI w Min w Mod w Ser : These correspond to the risk weights of the four categories of injury severity in the target node accident: no injury, minor injury, moderate injury, and severe injury, respectively, with values ​​increasing sequentially. If r i If the risk threshold constant for a road slice is constant, then the i-th road slice is considered a risky road slice, and the set of all risky slices is the road risk point, which is adjusted together with the content of the accident injury severity elements in the database.

[0094] Step 7: Calculate the overall road risk level and generate Output 2: Road Risk Measurement Index. The formula for overall road risk measurement is:

[0095]

[0096] In Equation 2, R: the quantitative index of the road risk being assessed.

[0097] n: The number of road segments in this road.

[0098] x i : The length of the i-th road slice.

[0099] l: The total length of the road.

[0100] r Max Risk value of the road segment with the highest risk in this road.

[0101] The above process realizes a method for assessing the road risk level of autonomous driving.

[0102] Next, based on steps one through five above, the evaluation method for the lane matching risk index of autonomous driving will be described in detail.

[0103] Step 8: Calculate the risk weights of the factors

[0104] Based on the Bayesian network risk quantitative assessment model, the sensitivity of nodes to the target is calculated. In Netica, the Bayesian network risk quantitative assessment model is used to perform sensitivity analysis with "Injury Severity" as the target through the "Network—Sensitivity to Findings" (data sensitivity analysis function). After the analysis, the influence weights of each factor on "Injury Severity" are output. After normalizing the target sensitivity, the risk weights of all factors can be obtained.

[0105] Step Nine: Calculate vehicle-road matching risk points. Match each element of each slice. If an element matches, assign a value of 1; if it does not match, assign a value of 0. The formula for calculating the vehicle-road slice matching risk index for each slice is as follows:

[0106]

[0107] In Equation 3, m v The risk index for matching vehicle-road slices in the v-th road slice.

[0108] q: The number of features in the v-th road slice.

[0109] e j In the v-th road slice, the j-th element is assigned a value of 1 if it matches the design operating conditions of the autonomous vehicle, and 0 if it does not match the design operating conditions of the autonomous vehicle.

[0110] w j The risk weight of the j-th element in the v-th road slice obtained in step f2.

[0111] When any road slice's vehicle-road slice matches the risk index m v When the risk threshold is less than the vehicle-road matching risk threshold, the slice of the road is determined to be the vehicle-road matching risk point of the vehicle. The set of all vehicle-road matching risk points in the road is output 3: vehicle-road matching risk point.

[0112] Step 10: Based on the risk weights of all elements, using the road element set of the proposed operating road and the design operating condition elements of the proposed operating vehicles, calculate and generate the output 4-vehicle-road matching risk assessment quantitative index. The calculation formula is as follows:

[0113]

[0114] In Equation 4, M: the vehicle-road matching risk index of the road.

[0115] p: The number of road segments in this road.

[0116] q: The number of elements in the road slice of this road.

[0117] x k : The length of the k-th road slice.

[0118] l: The total length of the road.

[0119] m kj In this road, for the k-th road slice, the j-th element's matching value is 1 if it matches the design operating conditions of the autonomous vehicle, and 0 if it does not match the design operating conditions of the autonomous vehicle.

[0120] w kj The weight of the j-th element in the k-th road slice of this road.

[0121] When the vehicle-road matching risk index M between the proposed operating road and the proposed autonomous vehicle is less than the vehicle-road matching risk threshold constant, it is determined that there is a risk in the autonomous vehicle operating on that road.

Claims

1. A Bayesian-based method for assessing the road risk of autonomous driving, wherein the method is implemented by processing and analyzing autonomous driving data using a computer and supporting software programs, characterized in that: The specific steps of the method are as follows: a1. Process the autonomous driving data to form a database including accident injury level, road type, road surface dryness and wetness, road speed limit, road condition, accident type, weather and lighting elements. Among them, the accident injury level element includes four items: no injury, minor injury, moderate injury and severe injury. b1. Based on the elements and content in the database in step a1, take the accident injury severity element as the target node and all other elements as nodes, and construct a Bayesian network risk qualitative assessment model in the supporting software program. c1. The Bayesian network risk qualitative assessment model is used to learn and train all accident data in the database to obtain the joint distribution of conditional probabilities of all nodes, i.e., each element, thus constructing the Bayesian network risk quantitative assessment model. d1. The autonomous driving roads to be evaluated are sliced ​​into multiple continuous road slices. The road slices are split in the following two ways: the same road type is split according to the unit mileage; different road types are split according to the differences in road characteristics. e1. Extract the road element set for each road slice according to the elements in the Bayesian network risk quantitative assessment model; f1. Based on the road element set of each road slice, conduct risk assessments on each road slice to form road risk points. Specifically, using the elements of the road slice as known factors, and according to the joint distribution of conditional probabilities in the Bayesian network risk quantitative assessment model constructed in step c1, calculate the probability of no injury, minor injury, moderate injury, and severe injury in the accident injury severity of the target node according to Bayesian network theory, and quantify the risk of the road slice. The risk quantification formula for any road slice is as follows: Formula 1, In Equation 1, Risk assessment results for the i-th road slice r NoIi For the i-th road slice, the probability of no damage is... r Mini Regarding the probability that the i-th road slice is slightly damaged, r Modi : The probability of moderate damage for the i-th road slice r Seri : The probability of severe damage to the i-th road slice w NoI w Min w Mod w Ser : These correspond to the risk weights of the four categories of injury severity in the target node accident: no injury, minor injury, moderate injury, and severe injury, respectively, with values ​​increasing sequentially. If r i >If the risk threshold constant for a road slice is constant, then the i-th road slice is considered a risky road slice, and the set of all risky slices is the road risk point; g1. Calculate the overall road risk of the autonomous driving road based on the risk level of any road slice obtained in step f1. The overall road risk quantification formula is as follows: Formula 2 In Equation 2, R: the quantitative index of the road risk being assessed. n: The number of road segments in this road. x i : The length of the i-th road slice. l: The total length of the road. r Max Risk value of the road segment with the highest risk in this road.

2. The Bayesian-based method for assessing the road risk of autonomous driving according to claim 1, characterized in that: In step e1, the element extraction process uses equipment equipped with a visual sensor to collect data, or it is collected by querying design and construction materials in conjunction with the site.

3. The Bayesian-based method for assessing the road risk of autonomous driving according to claim 1, characterized in that: In step a1, the data cleaning process includes deleting incomplete data, deleting duplicate data, deleting interfering data, deleting irrelevant data, and performing data mining to construct indirect elements.

4. A Bayesian-based method for assessing the risk index of autonomous vehicle-road matching, wherein the method is implemented by processing and analyzing autonomous driving data using a computer and supporting software programs, characterized in that: The specific steps of the method are as follows: a2. Process the autonomous driving data to form a database that includes accident injury level, road type, road surface dryness and wetness, road speed limit, road condition, accident type, weather and lighting elements. Among them, the accident injury level element includes four items: no injury, minor injury, moderate injury and severe injury. b2. Based on the elements and content in the database in step a2, take the accident injury severity element as the target node and all other elements as nodes, and construct a Bayesian network risk qualitative assessment model in the supporting software program. c2. The Bayesian network risk qualitative assessment model is used to learn and train all accident data in the database to obtain the joint distribution of conditional probabilities of all nodes, i.e., each element, thus constructing the Bayesian network risk quantitative assessment model. d2. The autonomous driving roads to be evaluated are sliced ​​into multiple continuous road slices. The road slices are split in the following two ways: the same road type is split according to the unit mileage; different road types are split according to the differences in road characteristics. e2. Extract the road element set for each road slice according to the elements in the Bayesian network risk quantitative assessment model; f2. Based on the Bayesian network risk quantitative assessment model in step c2, according to the Bayesian network technology theory, the sensitivity of all road element nodes and target nodes to the degree of accident damage in the Bayesian network risk quantitative assessment model is obtained. After normalizing the sensitivity, the risk weight of all elements to the degree of accident damage is obtained. g2. Calculate the vehicle-road segment matching risk index between each road slice and the vehicle-road segment to be operated by the autonomous vehicle. The calculation formula is as follows: Formula 3, In Equation 3, m v The risk index for matching vehicle-road slices in the v-th road slice. q: The number of features in the v-th road slice. e j In the v-th road slice, the j-th element is assigned a value of 1 if it matches the design operating conditions of the autonomous vehicle, and 0 if it does not match the design operating conditions of the autonomous vehicle. w j The risk weight of the j-th element in the v-th road slice obtained in step f2. When any road slice's vehicle-road slice matches the risk index m v When the value is less than the vehicle-road segment matching risk threshold constant, the segment of the road is determined to be the vehicle-road matching risk point of the vehicle. The set of all vehicle-road matching risk points in the road is the vehicle-road matching risk point. h2. Calculate the vehicle-road matching risk index between the proposed operating road and the proposed autonomous vehicles. The calculation formula is as follows: Equation 4, In Equation 4, M: the vehicle-road matching risk index of the road. p: The number of road segments in this road. q: The number of elements in the road slice of this road. x k : The length of the k-th road slice. l: The total length of the road. m kj In this road, for the k-th road slice, the j-th element's matching value is 1 if it matches the design operating conditions of the autonomous vehicle, and 0 if it does not match the design operating conditions of the autonomous vehicle. w kj The weight of the j-th element in the k-th road slice of this road. When the vehicle-road matching risk index M between the proposed operating road and the proposed autonomous vehicle is less than the vehicle-road matching risk threshold constant, it is determined that there is a risk in the autonomous vehicle operating on that road.

5. The Bayesian-based method for assessing the risk index of autonomous vehicle-road matching as described in claim 4, characterized in that: In step e2, the element extraction process uses equipment equipped with a visual sensor to collect data, or it is collected by querying design and construction materials in conjunction with the site.

6. The Bayesian-based method for assessing the risk index of autonomous vehicle-road matching as described in claim 4, characterized in that: In step a2, the data cleaning process includes deleting incomplete data, deleting duplicate data, deleting interfering data, deleting irrelevant data, and performing data mining to construct indirect elements.

Images