Automatic driving open test road environment risk degree determining and grading method and system
A road environment and automatic driving technology, which is applied in the traffic control system of road vehicles, traffic control system, traffic flow detection, etc., can solve the problems of insufficient accident data and insufficient data volume requirements, and achieve the effect of improving scientificity
Inactive Publication Date: 2020-09-15
TONGJI UNIV
0 Cites 6 Cited by
AI-Extracted Technical Summary
Problems solved by technology
In addition, at the current stage when autonomous driving is in the ascendant, there are very few accident data, whic...
Method used
[0148] In the above investigation process, data quality should be strictly controlled in three stages of design, implementation and result statistics. Control content includes data integrity, data authenticity, human error, data quality evaluation indicators, etc. Among them, the road user flow and traffic survey should ensure that the survey time covers the morning peak, evening peak and flat peak for at least one hour each; the road facility survey should ensure that the survey covers both sides of th...
Abstract
The invention discloses an automatic driving open test road environment risk degree determination and grading method and system, relates to the technical field of road traffic safety risk assessment and management,. The method includes: constructing various types of accident risk degree calculation models according to historical artificial driving road accident data; correcting each type of accident risk degree calculation model according to the vehicle accident data, and calculating each type of accident risk degree by utilizing the corrected each type of accident risk degree calculation model according to the road environment data and the traffic flow data; and determining the safety risk degree of the road environment according to the risk degrees of various accidents. By considering the risk causes of various accidents, the scientific calculation of the risk of the automatic driving open test road environment is realized.
Application Domain
Detection of traffic movementResources
Technology Topic
Road traffic safetyReal-time computing +7
Image
Examples
- Experimental program(1)
Example Embodiment
[0075] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative work shall fall within the protection scope of the present invention.
[0076] The purpose of the present invention is to provide a method and system for determining and grading the road environment risk of the automatic driving open test road, so as to realize the scientific calculation of the road environmental risk of the automatic driving open test road.
[0077] In order to make the above objectives, features and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.
[0078] Such as figure 1 As shown, the method for determining the road environment risk degree of the automatic driving open test provided by the present invention includes:
[0079] Step 101: According to historical manual driving road accident data, determine the road traffic accident risk cause and the influence coefficient corresponding to the road accident risk cause attribute value. The road traffic accident risk cause is used to determine the influencing factors of the road accident type.
[0080] Step 102: Construct various types of accident risk calculation models according to the road traffic accident risk cause and the influence coefficient corresponding to the attribute value of the road traffic accident risk cause.
[0081] Step 103: Acquire accident data of autonomous driving vehicles and accident data of manual driving vehicles.
[0082] Step 104: According to the accident data of autonomous vehicles and the accident data of manual driving vehicles, the risk calculation models of various types of accidents are modified to obtain the influence coefficient of the modified accident risk attribute value; specifically including: the automatic driving vehicle The accident data and manual driving vehicle accident data are input into the Bayesian network. The Bayesian network fits the automatic driving vehicle accident data and the manual driving vehicle accident data to obtain the vehicle accident data fitting value. When the vehicle accident data fitting value When the absolute value of the difference between the actual value of the accident data and the actual value of the accident data is less than or equal to the set threshold, the Bayesian network outputs the influence coefficient of the modified accident risk attribute value;
[0083] Step 105: Update the risk calculation models of various types of accidents according to the influence coefficients of the modified accident risk attribute values, and obtain the revised risk calculation models of various types of accidents. The revised risk calculation models of various types of accidents are automatic driving Various types of risk calculation models for vehicles.
[0084] Step 106: According to the road environment data and the traffic flow data, use the revised risk calculation model of various accidents to calculate the risk of various accidents.
[0085] Step 107: Determine the road environment safety risk degree according to the risk degree of each type of accident.
[0086] Among them, various types of accidents include: leaving the traffic lane accident, out of control and crashing into the opposing motor vehicle accident, overtaking and crashing into the opposing motor vehicle accident, intersection accident and road entrance accident. The calculation models of various types of accident risk are:
[0087] The risk of each type of accident = the probability of accident × the severity of the accident × the influence coefficient of vehicle speed in the road section × the influence coefficient of road section traffic flow × the influence coefficient of central partition type × the influence coefficient of weather environment × the influence coefficient of traffic composition.
[0088] Among them, the probability and severity of accidents are determined according to road environment data and traffic flow data, and the risk levels of various types of accidents are calculated according to the probability and severity of accidents.
[0089] In practical applications, the probability of accident and the severity of the accident are obtained by multiplying the influence coefficient of the attribute value of the road setting factor in the attribute value of the accident risk cause.
[0090] Determine the influence coefficient of the vehicle speed in the road section according to the vehicle speed in the road section.
[0091] Determine the influence coefficient of the road section traffic flow according to the section flow.
[0092] Determine the influence coefficient of the type of the central separation zone according to the type of the central separation zone.
[0093] Determine the influence coefficient of traffic composition according to the proportion of large vehicles in the traffic composition.
[0094] Among them, step 107 specifically includes:
[0095] The road environment safety risk degree includes road section safety risk degree, road access risk degree and road network risk degree.
[0096] According to the formula road section safety risk degree = Σ various types of accident risk degree, calculate the road section safety risk degree according to the formula
[0097] According to the formula
[0098] Such as figure 2 As shown, the present invention provides a system for determining the road environment risk of an open test of automatic driving, including:
[0099] The accident risk cause determination module 201 is used to determine the road traffic accident risk cause and the influence coefficient corresponding to the attribute value of the road traffic accident risk cause based on historical manual driving road accident data. The road traffic accident risk cause is used to determine road accidents Type of influencing factors.
[0100] The risk calculation model building module 202 is used to construct various types of accident risk calculation models based on the road traffic accident risk cause and the influence coefficient corresponding to the attribute value of the accident risk cause.
[0101] The vehicle accident data acquisition module 203 is used to acquire accident data of autonomous driving vehicles and accident data of manual driving vehicles.
[0102] The risk calculation model correction module 204 is used to correct various types of accident risk calculation models based on the automatic driving vehicle accident data, vehicle accident data and manual driving vehicle accident data, to obtain the modified impact coefficient of the accident risk cause attribute value , Specifically includes: inputting the accident data of autonomous vehicles, vehicle accident data and manual driving vehicle accident data into the Bayesian network, and the Bayesian network fits the accident data of autonomous vehicles and the accident data of manual driving vehicles to obtain the vehicle Accident data fitting value, when the absolute value of the difference between the vehicle accident data fitting value and the actual value of the accident data is less than or equal to the set threshold, the Bayesian network outputs the modified influence coefficient of the attribute value of the accident risk cause.
[0103] The risk calculation model update model 205 is used to update the risk calculation models of various types of accidents according to the influence coefficients of the modified accident risk attribute values to obtain the revised risk calculation models of various types of accidents, and the revised types of accidents The risk calculation model is a calculation model of various types of risk of autonomous vehicles.
[0104] The accident risk calculation module 206 is used to calculate the risk of various accidents by using the revised risk calculation model of various types of accidents based on road environment data and traffic flow data.
[0105] The road environment safety risk degree determination module 207 is used to determine the road environment safety risk degree according to the risk degree of various types of accidents.
[0106] Such as image 3 As shown, the present invention provides a road environment risk classification method for automatic driving open test, including:
[0107] Step 301: According to historical manual driving road accident data, determine the road traffic accident risk cause and the influence coefficient corresponding to the attribute value of the road accident risk cause. The road traffic accident risk cause is used to determine the influencing factors of the road accident type.
[0108] Step 302: Construct various types of accident risk calculation models according to the road traffic accident risk cause and the influence coefficient corresponding to the road traffic accident risk cause attribute value.
[0109] Step 303: Acquire accident data of autonomous driving vehicles and accident data of manual driving vehicles.
[0110] Step 304: According to the accident data of autonomous vehicles and the accident data of manual driving vehicles, correct the calculation models of various types of accident risk to obtain the influence coefficient of the attribute value of the modified accident risk; specifically including: The accident data and manual driving vehicle accident data are input into the Bayesian network. The Bayesian network fits the automatic driving vehicle accident data and the manual driving vehicle accident data to obtain the vehicle accident data fitting value. When the vehicle accident data fitting value When the absolute value of the difference between the actual value of the accident data and the actual value of the accident data is less than or equal to the set threshold, the Bayesian network outputs the modified influence coefficient of the attribute value of the accident risk cause.
[0111] Step 305: Update the risk calculation model of each type of accident according to the influence coefficient of the attribute value of the modified accident risk cause, and obtain the revised risk calculation model of each type of accident. The revised risk calculation model of each type of accident is automatic driving The calculation model of various types of accident risk of vehicles.
[0112] Step 306: According to the road environment data and the traffic flow data, calculate the risk degree of each type of accident by using the revised risk degree calculation model of each type of accident.
[0113] Step 307: According to the risk degree of various types of accidents and the preset road environment safety risk degree, a clustering method is adopted to obtain a road environment classification standard.
[0114] Step 308: Determine the road environment safety risk level according to the road environment classification standard and the risk levels of various types of accidents.
[0115] Among them, step 307 specifically includes:
[0116] According to the risk of various types of accidents and the preset road environment safety risk, the road environment safety risk value is used as the clustering object, and the K-Means clustering method is used to obtain the road environment classification standard.
[0117] Such as Figure 4 As shown, an automatic driving open test road environment risk grading system provided by the present invention includes:
[0118] The accident risk cause determination module 401 is used to determine the cause of road traffic accident risk and the influence coefficient corresponding to the attribute value of the road traffic accident risk cause based on historical manual driving road accident data. The cause of road traffic accident risk is used to determine road accidents Type of influencing factors.
[0119] The risk calculation model construction module 402 is used to construct various types of accident risk calculation models according to the road traffic accident risk cause and the influence coefficient corresponding to the attribute value of the accident risk cause.
[0120] Vehicle accident data acquisition module 403, used to acquire accident data of autonomous vehicles and manual driving vehicles
[0121] The risk calculation model correction module 404 is used to modify the risk calculation models of various types of accidents according to the accident data of autonomous vehicles and the accident data of manual driving vehicles to obtain the influence coefficient of the attribute value of the accident risk cause after correction , Specifically includes: inputting the accident data of autonomous vehicles, vehicle accident data and manual driving vehicle accident data into the Bayesian network, and the Bayesian network fits the accident data of autonomous vehicles and the accident data of manual driving vehicles to obtain the vehicle Accident data fitting value, when the absolute value of the difference between the vehicle accident data fitting value and the actual value of the accident data is less than or equal to the set threshold, the Bayesian network outputs the modified influence coefficient of the attribute value of the accident risk cause.
[0122] The risk calculation model update model 405 is used to update the risk calculation model of various types of accidents according to the influence coefficient of the attribute value of the modified accident risk cause, to obtain the revised risk calculation model of various types of accidents, and the revised types of accidents The risk calculation model is a calculation model of various types of risk of autonomous vehicles.
[0123] The accident risk calculation module 406 calculates the risk of various accidents by using the revised risk calculation model of various accidents based on the road environment data and traffic flow data.
[0124] The classification standard determination module 407 is used to obtain the road environment grade classification standard by using a clustering method according to the risk degree of various types of accidents and the preset road environment safety risk degree.
[0125] The risk level determination module 408 is configured to determine the road environment safety risk level according to the road environment level classification standard and the road environment safety risk level.
[0126] In addition, the present invention also provides a specific method for grading the road environment risk degree of automatic driving open test, such as Figure 5 As shown, including the following steps:
[0127] 1) Analyze the types of potential road accidents, refine the causes of accident risks for autonomous vehicles, and build various types of accident risk calculation models.
[0128] 11) Analyze the five types of accidents that may occur when a motor vehicle is driving on the road. They are: departure from the traffic lane, loss of control and collision with the opposing motor vehicle, collision with the opposing motor vehicle during overtaking, intersection accident and road access accident. The risk causes are shown in Table 1.
[0129] Table 1 Risk Cause Table
[0130]
[0131]
[0132] For different types of accidents, combining the characteristics of the accidents, clarify the related risk causes of each accident.
[0133] For example, the risk factors for leaving the traffic lane are lane width, curvature, signs and markings, road surface roughness, slope, expected speed, recommended speed, traffic flow, etc., non-intersection accidents do not need to consider the impact of intersection accidents.
[0134] 12) Build various types of accident risk calculation models based on the attribute value of each risk cause, as follows:
[0135] Risk of various types of accidents = probability of accident × severity of accident × influence coefficient of vehicle speed in the road section × influence coefficient of road section traffic flow × influence coefficient of central partition type × weather environment influence coefficient × influence coefficient of traffic composition
[0136] In the formula, the accident probability and the severity of the accident are calculated by cumulatively multiplying the influence coefficients corresponding to the attribute values of the road facility factors related to each accident.
[0137] The influence coefficient of vehicle speed in a road section is determined by the speed of vehicles in the road section. When the average speed of vehicles in the road section exceeds 120km/h, the influence coefficient is 1. The influence coefficient decreases as the average speed decreases.
[0138] The influence coefficient of section traffic flow is determined by section flow, and the influence coefficient increases with the increase of section flow.
[0139] The influence coefficient of the central separation zone type is related to the type of the central separation zone (passable/non-passable).
[0140] The influence coefficient of traffic composition mainly considers the influence of the proportion of large vehicles on the degree of risk. When the proportion of large vehicles is less than 10%, the influence coefficient is 1, and the influence coefficient increases as the proportion of large vehicles increases.
[0141] Weather influence coefficient: 1 under good weather, greater than 1 under bad weather, and the influence coefficient increases with the increase of the severity of the weather; the influence coefficient of the central partition type is only used to calculate part of the accident risk.
[0142] Through the calculation of the risk of various types of accidents, the road environment risk determination of the open test of automatic driving is completed.
[0143] 2) Based on the traditional manual driving vehicle accident data and the limited automatic driving accident data, the influence coefficient of the attribute value of each risk cause in the model is checked and revised; the Bayesian network is constructed according to the correlation of each risk cause. Input automatic driving accident data and manual driving accident data into the network for model training and learning. The accident data includes all accident types, corresponding risk-causing attribute values, and the influence coefficient of each risk-causing attribute value. According to the results of Bayesian network parameter learning, the influence coefficients of each risk cause in the model are checked and revised, and the Bayesian network outputs the revised influence coefficients of each risk cause;
[0144] 3) Calculate the road environment safety risk degree based on the road environment data and traffic flow data collected by field investigation;
[0145] 31) Investigation of road facility factors: Through field investigation, the main static factors affecting road safety, such as the number of lanes, road flatness, type of intermediate separation zone, curvature, slope, road access points, intersection conditions, etc., are recorded in detail. Circumstances, and the preparation of the influencing factor record table is conducive to the subsequent risk calculation.
[0146] 32) Road segment division: the same or similar continuous roads with the same or similar factors investigated in 31) are divided into the same road segment and numbered.
[0147] 33) Investigation of traffic factors: Vehicle speed and traffic flow are important dynamic factors that affect road safety risks. Vehicle speed can be determined according to the speed limit of the road section. Traffic flow needs to be investigated for each divided road section to obtain each The average daily traffic volume of the road section and the proportion of traffic types.
[0148] In the above survey process, the data quality should be strictly controlled from the three stages of design, implementation and result statistics. The control content includes data integrity, data authenticity, human error, data quality evaluation indicators, etc. Among them, road user traffic and traffic surveys should ensure that the survey time covers the morning peak, evening peak and peace peak for at least one hour; the road facility survey should ensure that the survey covers both sides of the road and cannot be interrupted or omitted; the road environment survey should ensure that the survey Simultaneity. For data surveys, priority is given to data obtained by reliable automated collection methods, such as videos, coils, etc. If there is a lack of corresponding data, on-site surveys are required. All on-site data collection must be carried out by three or more investigators who have received corresponding training at the same time to ensure the reliability of the investigation results. The survey results should be averaged.
[0149] 34) Formula for calculating road environment safety risk degree:
[0150] Road section safety risk = Σ various types of accident risk (2)
[0151] The calculation method of road access risk is shown in formula (3):
[0152]
[0153] Among them, n is the number of road sections in the road channel.
[0154] The calculation method of road network risk is shown in formula (4):
[0155]
[0156] Among them, n is the number of road sections in the road channel, and m is the number of road channels in the road network.
[0157] 4) The expert scoring method is used to collect data, the road environment grade classification standard is determined through clustering, and the road environment grade is determined based on the road environment safety risk.
[0158] 41) Design a questionnaire survey to collect the expert’s rating of the road environment to determine the road environment grading threshold. The questionnaire includes multiple sets of pictures and text descriptions of the road environment. The experts use a four-level grading method to rank, combined with the road environment information described in the questionnaire Give a rating result for each group of road environment (low risk/normal risk/higher risk/high risk).
[0159] 42) Calculate the risk degree of the road environment in the questionnaire according to the updated model in step 2), combine the rating results of all experts on the road environment, adopt the K-Means method, and use the obtained risk value as the cluster object to determine the risk classification threshold , And then clarify the risk range corresponding to each road environmental level.
[0160] 43) Match each calculated road safety risk degree into the road environment grade classification standard, and obtain the road environment risk grade corresponding to each road section. The road environment is divided into four levels, namely low risk (road environment class Ⅰ), general risk (road environment class Ⅱ), higher risk (road environment class Ⅲ), and high risk (road environment class Ⅳ).
[0161] A total of 11.1 kilometers of Moyu South Road-Anli Road-Bei'ande Road-Anzhi Road-Boyuan Road-Anhong Road-Antuo Road in the sunny state of Shanghai Jiading District was used as the research object to carry out road environment classification, and the research scope Such as Image 6 As shown, the specific implementation steps are as follows:
[0162] Step 1: Based on the five types of accidents, clarify the risk causes as shown in Table 1.
[0163] Step 2: Model revision. Construct a Bayesian network based on the correlation of each risk cause such as Figure 7 As shown, the autonomous driving accident data comes from the self-driving car operation accident report database established by the California Department of Motor Vehicles (DMV), and the manual driving accident data comes from the accident report sampling system established by the National Highway Traffic Safety Administration, referred to as CRSS (The Crash Report Sampling System). Refer to the existing literature of traditional manual driving accidents to determine the influence coefficient of each risk cause. According to the results of Bayesian network parameter learning, the revised results are shown in Table 2.
[0164] Table 2 Modified result table of risk-causing influence coefficient
[0165]
[0166]
[0167] Step 3: Carry out an investigation based on the factors determined in Step 1. According to the results of the investigation, the assessed road is divided into 53 sections. The results of the road segmentation are as follows: Figure 8 As shown, each road section is alternately represented by two colors of black and gray. After the investigation is completed, the road environment safety risk degree of each road section is calculated according to the model parameters modified in step 2.
[0168] Step 4: Based on the questionnaire data, cluster the risk values of each grade of road environment, determine the critical values between two adjacent grades as 3.5, 12.5, and 22.5 respectively, and use this value to determine the risk to road traffic safety as an index classification point Classify. The environmental classification standards defined based on the traffic safety risk assessment indicators are shown in Table 3.
[0169] Table 3 Standard Table of Environmental Registration Classification
[0170]
[0171]
[0172] Such as Picture 9 As shown, among all 53 road sections to be assessed, there are 40 road environment class I sections, accounting for 75%; 10 road environment class II sections, accounting for 19%; 3 road environment class III sections, accounting for 6% ; No road environment class IV section. Among them, the horizontal line in the figure represents the road environment category I section, the dashed line represents the road environment category II section, the double horizontal line represents the road environment category III section, and the unequal thick double horizontal line represents the road environment category IV section
[0173] The method and system for determining and grading the road environment risk of automatic driving open test provided by the present invention have the following obvious advantages:
[0174] 1. Originality: The selection and management of open test roads for automatic driving play an important role in the safe and orderly development of open test for automatic driving. It is of great significance to study the road environment classification of open test for automatic driving. Existing autopilot test results show that the cause of autopilot accidents is very different from that of traditional manual driving accidents, and the road facilities in the autopilot environment have also undergone great changes, so a more comprehensive analysis of the risks of autopilot accidents is needed Causes and calculation methods of risk.
[0175] 2. Completeness: The present invention is a complete automatic driving open test road environment risk classification method including accident type determination, influencing factor selection, factor investigation method, risk calculation method and risk classification method.
[0176] 3. Scientificity: The present invention fully considers the characteristics of the road environment required for automatic driving and automatic driving accidents, clarifies the types of potential accidents and the causes of accident risks, and based on the actual automatic driving open test results, the Bayesian network The influence coefficient of each cause is checked and revised. And use the expert scoring method to determine a reasonable road environment classification standard for automatic driving open test.
[0177] 4. Practicability: Comprehensive consideration, universal applicability, suitable for environmental classification of various types of roads including highways, urban roads, and rural roads.
[0178] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments, and the same or similar parts between the various embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant information can be referred to the description of the method part.
[0179] Specific examples are used in this article to describe the principles and implementation of the present invention. The description of the above examples is only used to help understand the method and core idea of the present invention; at the same time, for those skilled in the art, according to the present invention There will be changes in the specific implementation and scope of application. In summary, the content of this specification should not be construed as limiting the present invention.
PUM


Description & Claims & Application Information
We can also present the details of the Description, Claims and Application information to help users get a comprehensive understanding of the technical details of the patent, such as background art, summary of invention, brief description of drawings, description of embodiments, and other original content. On the other hand, users can also determine the specific scope of protection of the technology through the list of claims; as well as understand the changes in the life cycle of the technology with the presentation of the patent timeline. Login to view more.
Similar technology patents
Method for detecting water inlet noise of closestool water tank by using A-weighted sound power level measurement method
ActiveCN110243453Aimprove science
Owner:李文杰
Environmental influence coefficient calculation method and radioactivity acceptance criterion determination method
PendingCN112434405Aimprove science
Owner:CHINA NUCLEAR POWER ENG CO LTD
Method for detecting flushing noises of pedestal pan by sound power level parallelepiped measurement surface method
ActiveCN109752083Aimprove science
Owner:李文杰
Product information pushing method, electronic equipment and storage medium
PendingCN111199453Aimprove science
Owner:BANK OF CHINA
Method for detecting water intake noise of toilet water tank by A-weighted sound energy level measurement method
ActiveCN110160638Aimprove science
Owner:李文杰
Classification and recommendation of technical efficacy words
- improve science
Passenger flow counting method based on deep learning in vertical visual angle
Owner:SYSU CMU SHUNDE INT JOINT RES INST +1
Virtual item generation method and apparatus
Owner:SHENZHEN IDREAMSKY TECH
Visualized pronunciation teaching method and apparatus
InactiveCN101290720Aimprove science
Owner:TIANJIN XUNFEI INFORMATION TECH
Optical power therapentic equipment
InactiveCN101269257AImprove the level of informatizationimprove science
Owner:深圳市微创医学科技有限公司
Rainwater regulation and storage facility real-time scheduling method based on real-time meteorological information
ActiveCN110570126Aimprove scienceImprove running quality
Owner:FUZHOU UNIVERSITY