Safety assessment method and device for whole driving process of self-driving automobile and computer equipment

Abstract

The invention relates to a safety assessment method and device for the whole driving process of an autonomous vehicle and computer equipment, and the method comprises the steps: obtaining first driving data of a target vehicle, and obtaining second driving data of vehicles in a surrounding region of the target vehicle; and then inputting the first driving data and the second driving data into a risk assessment model to perform risk assessment on the target vehicle to obtain a risk assessment result, and finally inputting the risk assessment result into a safety assessment model to perform safety assessment on the target vehicle to obtain a safety assessment result. Wherein the risk assessment result comprises a driving risk, a collision risk and a damage risk. According to the method, the risk assessment model and the safety assessment model are utilized, and the driving risk, the collision risk and the damage risk are combined, so that the safety of the automatic driving vehicle is assessed more comprehensively and deeply, and the assessment accuracy can be improved.

Description

Technical Field

[0001] The present application relates to the field of autonomous driving technology, and in particular to a method, device, and computer equipment for evaluating the safety of an autonomous vehicle throughout its entire driving process. Background Art

[0002] As artificial intelligence matures, cities are becoming increasingly intelligent. Autonomous vehicles are a burgeoning industry within the AI field. While the continued advancement of autonomous driving technology may gradually reduce accident rates, existing data is insufficient to fully support safety comparisons between vehicles of different levels of autonomy in complex and critical scenarios.

[0003] Currently, safety assessments for autonomous vehicles primarily use the following three methods to compare the safety of vehicles at different levels of autonomy: First, target crash population analysis (CCP) matches autonomous driving features with specific accident types (rear-end collisions, ghosting scenarios) to estimate the effectiveness of autonomous driving technology in preventing potential collisions. Second, traffic flow simulation constructs a virtual traffic environment to assess the impact of autonomous vehicles of varying levels and penetration rates on overall traffic flow dynamics. Third, driving simulator-based research evaluates human-vehicle interactions in specific scenarios.

[0004] However, the safety assessment methods for autonomous vehicles suffer from inaccurate assessments. Summary of the Invention

[0005] Based on this, it is necessary to provide a method, device and computer equipment for evaluating the safety of an autonomous vehicle during its entire driving process, which can improve the accuracy of the safety evaluation of the autonomous vehicle during its entire driving process, in response to the above technical problems.

[0006] In a first aspect, the present application provides a method for evaluating the safety of an autonomous vehicle during its entire driving process, the method comprising:

[0007] Acquiring first driving data of a target vehicle, and acquiring second driving data of vehicles in an area surrounding the target vehicle;

[0008] Inputting the first driving data and the second driving data into a risk assessment model to perform a risk assessment on the target vehicle to obtain a risk assessment result; the risk assessment result includes driving risk, collision risk, and damage risk;

[0009] The risk assessment results are input into the safety assessment model to perform safety assessment on the target vehicle and obtain the safety assessment results.

[0010] In some embodiments, the risk assessment model includes a driving risk assessment sub-model, a collision risk assessment sub-model, and a damage risk assessment sub-model. The first driving data and the second driving data are input into the risk assessment model to perform a risk assessment on the target vehicle, and a risk assessment result is obtained, including:

[0011] Inputting the first driving data into the driving risk assessment sub-model for assessment to obtain driving risk;

[0012] Inputting the first driving data and the second driving data into a collision risk assessment sub-model for assessment to obtain a collision risk;

[0013] The first driving data and the second driving data are input into the damage risk assessment sub-model for assessment to obtain the damage risk.

[0014] In some embodiments, the driving risk assessment sub-model includes a weight analysis sub-model and a first calculation sub-model. The first driving data is input into the driving risk assessment sub-model for evaluation to obtain the driving risk, including:

[0015] Inputting the first driving data into the weight analysis sub-model to obtain the weights of the target factors; the target factors include vehicle speed factor, acceleration factor, lane departure distance factor and environmental factor;

[0016] The weights of the target factors are input into the first calculation sub-model to obtain the driving risk.

[0017] In some embodiments, the collision risk assessment sub-model includes an analysis sub-model and a second calculation sub-model. The first driving data and the second driving data are input into the collision risk assessment sub-model for evaluation to obtain the collision risk, including:

[0018] Inputting the first driving data and the second driving data into the analysis sub-model to obtain a longitudinal collision risk and a lateral collision risk;

[0019] The longitudinal collision risk and the lateral collision risk are input into the second calculation sub-model to obtain the collision risk.

[0020] In some embodiments, the method for obtaining the damage risk assessment sub-model includes:

[0021] Acquiring collision velocity sample data and other sample data of the vehicle under a collision condition; the other sample data includes at least one of initial collision condition sample data, occupant physiological parameter sample data, and occupant posture sample data;

[0022] High-dimensional processing is performed on other sample data to obtain processed other sample data, and an initial damage risk assessment sub-model is trained based on the processed other sample data to obtain a damage risk assessment sub-model.

[0023] In some embodiments, the method further comprises:

[0024] Obtain a safety assessment scale; the safety assessment scale includes multiple driving factors, each driving factor includes multiple driving safety questions, and the scores corresponding to each driving safety question;

[0025] A safety assessment model is constructed based on the safety assessment scale and driving risk parameters, collision risk parameters and injury risk parameters.

[0026] In some embodiments, a safety assessment model is constructed based on the safety assessment scale and the driving risk parameter, the collision risk parameter, and the injury risk parameter, including:

[0027] Construct a driving factor loading matrix based on the safety assessment scale;

[0028] Determine the load value corresponding to each driving factor according to the driving factor load matrix;

[0029] Determine the weight of the driving risk parameter, the weight of the collision risk parameter, and the weight of the injury risk parameter based on the load value corresponding to each driving factor and the score corresponding to each driving safety question;

[0030] A safety assessment model is constructed based on the weights of driving risk parameters, collision risk parameters, injury risk parameters, driving risk parameters, collision risk parameters and injury risk parameters.

[0031] In a second aspect, the present application also provides a device for evaluating the safety of an autonomous vehicle during its entire driving process, the device comprising:

[0032] an acquisition module, configured to acquire first driving data of a target vehicle and second driving data of vehicles in an area surrounding the target vehicle;

[0033] a risk assessment module, configured to input the first driving data and the second driving data into a risk assessment model to perform a risk assessment on the target vehicle and obtain a risk assessment result; the risk assessment result includes driving risk, collision risk, and damage risk;

[0034] The safety assessment module is used to input the risk assessment results into the safety assessment model to perform a safety assessment on the target vehicle and obtain a safety assessment result.

[0035] In a third aspect, the present application also provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the method for safety assessment of the entire driving process of an autonomous driving vehicle as described in any embodiment of the first aspect above are implemented.

[0036] In a fourth aspect, the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the method for safety assessment of the entire driving process of an autonomous driving vehicle as described in any embodiment of the first aspect above.

[0037] In a fifth aspect, the present application also provides a computer program product, which includes a computer program. When the computer program is executed by a processor, it implements the method for safety assessment of the entire driving process of an autonomous driving vehicle as described in any embodiment of the first aspect above.

[0038] The above-mentioned method, apparatus, and computer device for assessing the safety of an autonomous vehicle throughout its entire driving process obtains first driving data of a target vehicle and second driving data of vehicles in the target vehicle's surrounding area. The method then inputs the first and second driving data into a risk assessment model to perform a risk assessment on the target vehicle, obtaining a risk assessment result. Finally, the risk assessment result is input into a safety assessment model to perform a safety assessment on the target vehicle, obtaining a safety assessment result. The risk assessment result includes driving risk, collision risk, and injury risk. In this method, the risk assessment model and the safety assessment model are combined with driver behavior data and vehicle dynamics data in normal and hazardous traffic scenarios to extend the safety assessment from hazardous scenarios forward to the normal driving process and backward to collisions and occupant injuries. This extends the safety benefit assessment from traditional driving risk quantification to collision risk quantification and occupant injury risk assessment, achieving a more comprehensive and in-depth safety assessment of the autonomous vehicle, thereby improving the accuracy of the assessment. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 is a diagram of the internal structure of a computer device in some embodiments;

[0040] Figure 2 This is a flowchart of a method for assessing the safety of an autonomous vehicle during its entire driving process in some embodiments;

[0041] Figure 3 This is a second flow chart of a method for assessing the safety of an autonomous vehicle during its entire driving process in some embodiments;

[0042] Figure 4 This is a third flow chart of a method for assessing the safety of an autonomous vehicle during its entire driving process in some embodiments;

[0043] Figure 5 This is a fourth flow chart of a method for assessing the safety of an autonomous vehicle during its entire driving process in some embodiments;

[0044] Figure 6This is a fifth flowchart of a method for assessing the safety of an autonomous vehicle during its entire driving process in some embodiments;

[0045] Figure 7 6. A flowchart of a method for assessing the safety of an autonomous vehicle during its entire driving process, according to some embodiments;

[0046] Figure 8 1 is a structural block diagram of a device for evaluating the safety of an autonomous vehicle during its entire driving process in some embodiments. DETAILED DESCRIPTION

[0047] In the embodiments of this application, the term "and / or" describes the association relationship between associated objects, indicating that three relationships can exist. For example, A and / or B can represent three situations: A exists alone, A and B exist simultaneously, and B exists alone. The character " / " generally indicates that the associated objects are in an "or" relationship.

[0048] In the embodiments of the present application, the term "plurality" refers to two or more than two, and other quantifiers are similar.

[0049] In the embodiments of the present application, the term "at least one" means one or more. For example, at least one of A, B and C can mean the following six situations: A exists alone, B exists alone, C exists alone, A and B exist at the same time, A and C exist at the same time, B and C exist at the same time, and A, B and C exist at the same time.

[0050] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0051] Globally, road traffic accidents cause 1.19 million deaths annually, with passenger fatalities accounting for the largest share, 29% of all road accident deaths. The road safety situation is severe. Improving vehicle safety and reducing road traffic accidents and casualties are major social issues that need to be addressed urgently. Vehicle safety and the protection of road users are pillars of research for improving road traffic safety. Road traffic scenarios are complex and ever-changing. The dynamic interaction of the three elements of "driver-vehicle-road" constitutes a highly dynamic, strongly coupled, and nonlinear generalized dynamic system. Safety-critical scenarios, distinct from normal driving scenarios, are those in which traffic conflict sources are present and could potentially escalate into a collision. These critical scenarios play a key role in safety risks. When a critical scenario occurs, the driver goes through three key processes: hazard perception, driving decision-making, and collision avoidance execution. These processes control the interaction between the vehicle and the hazard source and influence whether a collision occurs. If a collision is not avoided, the human body is subjected to impact loads and can sustain injuries. As vehicles become increasingly intelligent, autonomous vehicles are believed to significantly improve overall road safety by significantly reducing accidents caused by human driver error, distraction, or fatigue. However, existing data indicates that the actual safety performance of autonomous vehicles has yet to significantly surpass that of conventional vehicles. According to 2021 statistics from the U.S. National Highway Traffic Safety Administration (NHTSA), autonomous vehicles were involved in an average of 9.1 crashes per million miles traveled, compared to 4.2 for conventional vehicles. This comparison highlights the safety deficiencies of autonomous driving technology at this stage. Furthermore, assessing the safety of autonomous vehicles remains challenging. The scarcity of safety-critical scenarios is a key issue, requiring large-scale real-world accident data for analysis and validation. However, existing real-world accident data has limitations in the following aspects: Comparing the safety of vehicles of different levels of autonomy fails to fully account for the relative consistency of road conditions, accident types, pre-accident vehicle conditions, and the urgency of traffic scenarios. Consequently, some safety conclusions cannot be directly compared due to highly inconsistent prior assumptions. Furthermore, vehicles of different levels of autonomous driving differ significantly in terms of number, testing sites, operating conditions, and accident rate evaluation criteria, making it difficult to compare their safety under similar or comparable operating conditions. While the continued development of autonomous driving technology may gradually reduce its accident rate, existing data is insufficient to fully support safety comparisons between vehicles of different levels of autonomous driving in complex and critical scenarios. Existing research primarily compares the safety of vehicles of different levels of autonomous driving through the following three methods: First, Target Crash Population analysis, which matches autonomous driving functions with specific accident types (rear-end collisions, "ghosting" scenarios) to estimate the effectiveness of autonomous driving technology in preventing potential collisions.The second is traffic flow simulation, which constructs a virtual traffic environment to evaluate the impact of autonomous vehicles of different levels and penetration rates on overall traffic flow dynamics. Third, research based on driving simulators can fully consider the behavioral characteristics of the driver and assess the human-vehicle interaction in specific scenarios in real time. Although existing methods are effective in certain scenarios, they still have significant limitations in data collection and safety benefit comparison in dangerous scenarios. First, analysis methods based on target collision populations are only applicable to low-level autonomous driving systems, such as simple AEB test scenarios, and have limited applicability to complex traffic scenarios. Methods based on traffic flow simulations rely too heavily on surrogate safety metrics (SSMs) calculated from physical parameters to assess risk, such as the distance to collision (TTC) between two vehicles. This indirect assessment method has limitations in accuracy, and the metric is not linearly related to the degree of collision. Research based on driving simulators often lacks collision events and focuses too much on the interaction between the driver and the vehicle, thereby ignoring the potential consequences of accidents. In summary, the current safety assessment methods for the entire driving process of autonomous vehicles have the problem of inaccurate assessment. There is an urgent need for a new technical solution that can effectively collect data on the entire process of autonomous vehicles in dangerous scenarios, including driving behavior, collision conditions, and occupant injury risks, and propose a unified real-time safety benefit assessment model that integrates driving risk, collision risk, and occupant injury risk, so as to objectively and reasonably evaluate the safety of future road traffic scenarios.

[0052] This application provides a method for assessing the safety of autonomous vehicles throughout their entire driving process, relating to the field of road traffic safety. By integrating driving risk, collision risk, and occupant injury risk in critical driving scenarios, an innovative model for real-time assessment of the safety benefits of autonomous vehicles has been developed. This model extends the safety assessment of autonomous vehicles in normal driving scenarios to critical driving scenarios (i.e., hazardous driving conditions). It can incorporate collision risk and occupant injury risk in critical driving scenarios into the quantification of overall road traffic safety, thereby comprehensively considering driving risk, collision risk, and occupant injury risk to assess the safety of autonomous vehicles. The following examples will specifically illustrate the method for assessing the safety of autonomous vehicles throughout their entire driving process described in this application.

[0053] The safety assessment method for the entire driving process of an autonomous vehicle provided in the embodiment of the present application can be applied to Figure 1 In the computer device shown in FIG. 1 , the computer device may be an autonomous driving system of an autonomous vehicle. The internal structure diagram thereof may be as follows: Figure 1As shown, the computer device includes a processor, memory, an input / output interface, a communication interface, a display unit, and an input device. The processor, memory, and input / output interface are connected via a system bus, while the communication interface, display unit, and input device are connected to the system bus via the input / output interface. The processor of the computer device provides computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operating system and computer programs stored in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals via wired or wireless means, and the wireless means may be implemented via Wi-Fi, a mobile cellular network, NFC (near-field communication), or other technologies. When executed by the processor, the computer program implements a method for assessing the safety of an autonomous vehicle throughout its entire driving process. The display unit of the computer device is used to produce visual images and may be a display screen, a projection device, or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad set on the computer device casing, or an external keyboard, touchpad or mouse.

[0054] Those skilled in the art will understand that Figure 1 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0055] In some embodiments, as Figure 2 As shown in the figure, a safety assessment method for the whole process of autonomous driving vehicle is provided. Figure 1 The computer device in the example is used to illustrate the process, including the following steps:

[0056] S201 : Acquire first driving data of a target vehicle, and acquire second driving data of vehicles in an area surrounding the target vehicle.

[0057] The target vehicle may be an autonomous vehicle. According to the Society of Automotive Engineers (SAE), autonomous driving is divided into six levels (Level 0 to 5) based on the degree of automation and the need for driver involvement. Level 0 means the vehicle is fully driver-controlled, with the system providing only alerts or assistance functions, such as collision warnings or blind-spot monitoring. Level 5 means the vehicle can drive completely autonomously under all conditions, without driver involvement or intervention, and may not even have a steering wheel or pedals. The target vehicle may be any of Levels 0 to 5.

[0058] The first driving data is the target vehicle's driving data, including at least one of vehicle-level data, road-level data, occupant-level data, and restraint system-level data. Vehicle-level data includes at least one of the vehicle's mass, dimensions (length and width), stiffness, driving speed, driving acceleration, heading angle, collision velocity, collision angle, overlap ratio, and position. Road-level data includes at least one of the relative distance between the two vehicles, relative speed, road friction coefficient, collision object type (such as a utility pole, guardrail, or vehicle type), lane width, road curvature, and lane coordinates. Occupant-level data includes accelerator pedal force, brake pedal force, steering wheel angle, steering wheel torque, and at least one of occupant height, occupant weight, seat fore-aft position, occupant fore-aft tilt, and side-to-side tilt. Restraint system-level data includes at least one of the driver's head airbag status (whether equipped or deployed), knee airbag status, and seatbelt force limiter. The second driving data is the driving data of vehicles in the target vehicle's surrounding area, including at least one of vehicle-level data, road-level data, occupant-level data, and restraint system-level data.

[0059] In this embodiment of the present application, the target vehicle's mass, dimensions, width, rigidity, and seatbelt force limiter can be pre-stored in the vehicle's electronic control unit (ECU) or database. When assessing the safety of the target vehicle, the computer device can utilize vehicle-level data of the target vehicle, data on the road the target vehicle is traveling on, driving data on the occupants of the target vehicle, and data on the restraint systems installed on the target vehicle. Furthermore, the computer device can obtain secondary driving data of surrounding vehicles via a network.

[0060] Specifically, when the computer obtains vehicle-level data of the target vehicle, it can read the target vehicle's mass, size, stiffness and other data from the local electronic control unit (ECU) or database; obtain the target vehicle's driving speed through the speed sensor on the target vehicle; obtain the target vehicle's driving acceleration through the acceleration sensor; obtain the target vehicle's heading angle through the gyroscope or inertial measurement unit (IMU); when the target vehicle collides, it can obtain the collision speed through the speed sensor, obtain the collision angle through the visual sensor or lidar or collision sensor, and obtain the target vehicle's volume and area and other related features through the camera to calculate the overlap rate.

[0061] When the computer obtains the road-level data on the target vehicle, it can use lidar or millimeter-wave radar to measure the relative distance and relative speed between the vehicle and surrounding vehicles in real time by emitting electromagnetic waves and receiving reflected waves; obtain the road friction coefficient through tire force sensors, or obtain road humidity and temperature information through roadside sensors to calculate the road friction coefficient; when the target vehicle collides, the type of collision object can be obtained through the camera; lane line width, road curvature and lane line coordinate position can be obtained through GPS technology or other high-precision map technologies.

[0062] When acquiring driving data from the occupants of the target vehicle, the computer can obtain accelerator and brake pedal forces through pressure sensors on the accelerator and brake pedals, and obtain steering wheel angle and steering wheel torque through a steering wheel angle sensor and torque sensor. It can use in-vehicle cameras (such as those located on the roof or above the center console) to identify the positions of key points such as the occupant's head and feet, and estimate the occupant's height based on the camera's installation position, angle, and calibration parameters. Furthermore, the computer can use pressure sensors installed on the seats of the target vehicle to detect pressure distribution at different locations when the occupant sits in the seat. By analyzing the pressure distribution data, the occupant's weight can be determined. It should be noted that occupants include both the driver and passengers. The fore-and-aft position of the seat can be acquired through position sensors on the seats of the target vehicle. An angle sensor installed at the junction of the seat backrest and the seat cushion can accurately measure the rotation angle of the seat backrest relative to the cushion. When the occupant adjusts the fore-and-aft tilt of the seat backrest, the tilt sensor can determine the tilt angle of the seat backrest by detecting the angle between the direction of gravity and the sensor axis, thereby obtaining the occupant's fore-and-aft tilt angle and the left-and-right tilt angle.

[0063] When the computer obtains the restraint system-level data equipped on the target vehicle, it monitors the status of the head airbag and knee airbag in real time through the target vehicle's airbag control system, including whether the airbag is equipped and whether the airbag is deployed; and reads the target vehicle's seat belt force limit value from the local electronic control unit (ECU) or database.

[0064] S202: Input the first driving data and the second driving data into a risk assessment model to perform risk assessment on the target vehicle to obtain a risk assessment result.

[0065] Among them, the risk assessment model can be a neural network model, a machine learning model, or a mathematical model. The risk assessment results include driving risk, collision risk, and injury risk; optionally, the risk assessment results can also include the risk level or risk degree corresponding to each of the driving risk, collision risk, and injury risk. Driving risk assessment helps to discover in advance the problems that may exist in the target vehicle during normal driving, such as speeding, illegal lane changes, etc.; collision risk assessment focuses on predicting the possibility of a vehicle collision during driving and the degree of damage that may be caused to the target vehicle after a collision; injury risk assessment further considers the degree of injury that may be caused to the occupants after a collision.

[0066] In an embodiment of the present application, a machine learning or deep learning algorithm can be used to construct an initial risk assessment model in advance. A large amount of driving sample data (including data from normal driving, data from dangerous driving conditions, data from near-collision situations, data from collision avoidance attempts, and data after a collision) is then acquired. The driving sample data is then labeled, with the labeled information including the degree of driving risk, collision risk, and injury risk. The initial risk assessment model is then trained based on the driving sample data, and the model parameters are adjusted so that the model can accurately predict risks in different driving scenarios and learn the relationship between different driving data and risks, thereby obtaining a risk assessment model. After the computer device obtains the first and second driving data based on the above steps, the first and second driving data can be input into the risk assessment model. The risk assessment model then performs calculations and analyses based on the input data, and uses the learned rules and patterns to assess the driving risk, collision risk, and injury risk of the target vehicle, thereby obtaining a risk assessment result. Optionally, after the computer device obtains the first driving data and the second driving data based on the above steps, it can perform cleaning, filtering, and calibration on the first driving data and the second driving data to remove noise and errors, thereby obtaining preprocessed first driving data and preprocessed second driving data, and then input the preprocessed first driving data and the preprocessed second driving data into a risk assessment model to obtain a risk assessment result.

[0067] S203: Input the risk assessment result into the safety assessment model to perform safety assessment on the target vehicle to obtain a safety assessment result.

[0068] The safety assessment results include the safety status and safety level of the target vehicle. The safety assessment model can be a neural network model, a machine learning model, or a mathematical model.

[0069] In an embodiment of the present application, a machine learning or deep learning algorithm can be used in advance to construct an initial safety assessment model. A large amount of risk sample data is then acquired (including the target vehicle's driving risk, collision risk, and damage risk data during normal driving; the target vehicle's driving risk, collision risk, and damage risk data during dangerous driving conditions; the target vehicle's driving risk, collision risk, and damage risk data when a collision is imminent; the target vehicle's driving risk, collision risk, and damage risk data during collision avoidance; and the target vehicle's driving risk, collision risk, and damage risk data after a collision). The risk sample data is then labeled, with the labeled information including the vehicle's safety status and safety level. The initial safety assessment model is then trained based on the risk sample data, and the model's parameters are adjusted so that the model can accurately predict risks in different driving scenarios and learn the relationship between different driving data and safety status and safety level, thereby obtaining a safety assessment model. After the computer device obtains a risk assessment result based on the above steps, the risk assessment result can be input into the safety assessment model. The safety assessment model performs calculations and analyses based on the input data, and uses the learned rules and patterns to assess the target vehicle's safety status and safety level to obtain a safety assessment result. It should be noted that the methods described in the embodiments of the present application can be applied to autonomous vehicles during testing to achieve a safety assessment of the entire driving process of the autonomous vehicle during testing, and the autonomous vehicle can be subsequently operated and managed based on the safety assessment results. Optionally, the methods described in the embodiments of the present application can also be applied to autonomous vehicles in real life to achieve a safety assessment of the entire driving process of the autonomous vehicle in real life, and the traffic management department can subsequently monitor and manage traffic safety based on the safety assessment results. Optionally, the methods described in the embodiments of the present application can also be applied to other scenarios, and the specific scenarios can be determined according to actual needs.

[0070] The embodiment of the present application provides a method for assessing the safety of an autonomous vehicle during its entire driving process. The method obtains first driving data of a target vehicle and second driving data of vehicles in the surrounding area of the target vehicle, then inputs the first driving data and the second driving data into a risk assessment model to perform a risk assessment on the target vehicle to obtain a risk assessment result. Finally, the risk assessment result is input into a safety assessment model to perform a safety assessment on the target vehicle to obtain a safety assessment result. The risk assessment result includes driving risk, collision risk, and injury risk. In the above method, the risk assessment model and the safety assessment model are combined with the driver behavior data and vehicle dynamics data in normal traffic scenarios and dangerous traffic scenarios to extend the safety assessment from dangerous scenarios to the normal driving process and backward to collisions and occupant injuries. The safety benefit assessment is extended from traditional driving risk quantification to collision risk quantification and occupant injury risk assessment, thereby achieving a more comprehensive and in-depth assessment of the safety of autonomous vehicles, thereby improving the accuracy of the assessment.

[0071] In some embodiments, the above risk assessment model includes a driving risk assessment sub-model, a collision risk assessment sub-model and a damage risk assessment sub-model. On this basis, a specific implementation method for risk assessment of a target vehicle is provided, such as Figure 3 As shown, the above-mentioned step S202 of “inputting the first driving data and the second driving data into the risk assessment model to perform risk assessment on the target vehicle to obtain a risk assessment result” includes:

[0072] S301: Input the first driving data into a driving risk assessment sub-model for assessment to obtain driving risk.

[0073] The risk assessment model includes driving risk assessment sub-models, collision risk assessment sub-models, and damage risk assessment sub-models. It considers the global stability of vehicle operation and quantifies potential abnormal conditions during driving. Driving risks primarily include vehicle deviation from the normal driving trajectory, loss of stability, or approaching a critical state.

[0074] In an embodiment of the present application, a machine learning or deep learning algorithm can be pre-used to construct an initial driving risk assessment sub-model. A large amount of driving sample data (including data from normal driving, data from dangerous driving conditions, data from near-collision situations, data from collision avoidance attempts, and data after a collision) is then acquired. The driving sample data is then labeled, with the labeled information including driving risk. The initial driving risk assessment sub-model is then trained based on the driving sample data, and the model parameters are adjusted to enable the model to accurately predict driving risk in different driving scenarios. The model learns the relationship between different driving data and driving risk, thereby obtaining a driving risk assessment sub-model. After obtaining the first driving data, the computer device can input the first driving data into the driving risk assessment sub-model. The driving risk assessment sub-model then performs calculations and analyses based on the input data, and utilizes the learned rules and patterns to assess the driving risk of the target vehicle, thereby obtaining the driving risk.

[0075] Optional, such as Figure 4 As shown, the driving risk assessment sub-model includes a weight analysis sub-model and a first calculation sub-model. The step S301 of "inputting the first driving data into the driving risk assessment sub-model for assessment to obtain the driving risk" includes:

[0076] S3011: Input the first driving data into the weight analysis sub-model to obtain the weight of the target factor.

[0077] The driving risk assessment submodel includes a weight analysis submodel and a first calculation submodel. Target factors include vehicle speed, acceleration, lane departure distance, and environmental factors. The weight analysis submodel is fitted using the least squares method.

[0078] In an embodiment of the present application, a weight analysis sub-model can be constructed in advance based on the least squares method. After obtaining the first driving data, the computer device can input information such as the speed, acceleration, whether an accident has occurred, and whether the target vehicle is in a dangerous state in the first driving data into the weight analysis sub-model to obtain the weight of the target factor.

[0079] S3012: Input the weight of the target factor into the first calculation sub-model to obtain the driving risk.

[0080] In the embodiment of the present application, a first calculation sub-model may be constructed in advance based on the vehicle speed factor, the acceleration factor, the lane departure distance factor, the environmental factor, and the weights corresponding to each factor, as follows:

[0081]

[0082]

[0083] in, Indicates driving risk; represents the standard deviation of vehicle speed fluctuation (i.e., vehicle speed factor); represents the lateral acceleration (i.e., acceleration factor); Indicates the lane lateral deviation distance (i.e., lane deviation distance factor), which is determined by the vehicle position coordinates and lane line coordinates; Indicates environmental complexity (i.e., environmental factors); Indicates the weight coefficient of each target factor; represents the traffic flow density; Indicates the road curvature; Indicates weather conditions (W=1 indicates sunny, W=1.5 indicates light rain or snow, and W=2 indicates heavy rain, heavy snow, or dense fog).

[0084] S302 : Input the first driving data and the second driving data into a collision risk assessment sub-model for assessment to obtain a collision risk.

[0085] Among them, the collision risk is determined by the interaction between the vehicle and other vehicles in the environment during the vehicle's driving process.

[0086] In an embodiment of the present application, a machine learning or deep learning algorithm can be pre-used to construct an initial collision risk assessment sub-model. A large amount of driving sample data (including data on impending collisions, collision avoidance attempts, and post-collision data) is then acquired. The driving sample data is then labeled with information including collision risk. The initial collision risk assessment sub-model is then trained based on the driving sample data, and the model parameters are adjusted to enable the model to accurately predict collision risks in different driving scenarios. The model learns the relationship between different driving data and driving risks, thereby obtaining a collision risk assessment sub-model. After obtaining the first and second driving data, the computer device can input the first and second driving data into the collision risk assessment sub-model. The collision risk assessment sub-model then performs calculations and analyses based on the input data, utilizing the learned rules and patterns to assess the collision risk of the target vehicle and obtain a collision risk.

[0087] Optional, such as Figure 5 As shown, the collision risk assessment sub-model includes an analysis sub-model and a second calculation sub-model. The step S302 of "inputting the first driving data and the second driving data into the collision risk assessment sub-model for assessment to obtain the collision risk" includes:

[0088] S3021: Input the first driving data and the second driving data into an analysis sub-model to obtain a longitudinal collision risk and a lateral collision risk.

[0089] The collision risk assessment sub-model includes an analysis sub-model and a second calculation sub-model.

[0090] In the embodiment of the present application, an analysis sub-model can be pre-constructed, specifically as follows: based on the classification of the many environmental vehicles around the target vehicle, environmental vehicles whose distance to the ego vehicle is less than a preset distance (such as 100 meters) are defined as "dangerous environmental vehicles", otherwise they are defined as "safe environmental vehicles". For each dangerous environmental vehicle, the longitudinal collision risk and lateral collision risk between it and the ego vehicle are determined. Specifically, the longitudinal collision risk takes TTC as an example, which indicates the time remaining before the ego vehicle and the dangerous environmental vehicle maintain the current driving state (vehicle speed, heading angle). The lateral collision risk is calculated by the lateral distance between the ego vehicle and the dangerous vehicle ( ). The specific mathematical expression of longitudinal collision risk is as follows:

[0091]

[0092] in, represents the relative distance between the vehicle and an environment vehicle at time t, It represents the relative speed between the ego vehicle and the environment vehicle in the direction of ego vehicle's travel at time t.

[0093] After obtaining the first driving data and the second driving data, the computer device can input the first driving data and the second driving data into the analysis sub-model. The first driving data and the second driving data can be used to objectively quantify the safety at a certain moment and obtain the longitudinal collision risk and lateral collision risk of the target vehicle.

[0094] S3021: Input the longitudinal collision risk and the lateral collision risk into the second calculation sub-model to obtain the collision risk.

[0095] In the embodiment of the present application, a second calculation sub-model can be constructed in advance based on the longitudinal collision risk and the lateral collision risk, specifically as follows: when the time threshold is less than a pre-set time threshold (such as TTC < 2 seconds), the longitudinal collision risk is taken as the inverse and the duration less than the threshold is integrated to obtain the collision risk. ,as follows:

[0096]

[0097]

[0098] Among them, RT represents the road type; Indicates the weight coefficient of each factor; t indicates the duration of TTC being less than the pre-set threshold; Indicates the risk of collision, The parameters in the model can be based on actual accident statistics of cars with different levels of autonomous driving (crash rate as model output, detailed collision conditions, mileage and other parameters as model input), and confirmed using fitting methods, such as the least squares method.

[0099] After obtaining the longitudinal collision risk and the lateral collision risk, if the longitudinal collision risk is less than 2 seconds, the computer device will use the longitudinal collision risk and the lateral collision risk to obtain the collision risk of the target vehicle. If the longitudinal collision risk is not less than 2 seconds, it is determined that there is no collision risk for the target vehicle, that is, the collision risk of the target vehicle is zero.

[0100] S303: Input the first driving data and the second driving data into a damage risk assessment sub-model for assessment to obtain damage risk.

[0101] Among them, injury risk indicates the severity of occupant injury and is an important basis for safety benefit assessment. The core protection object of autonomous driving vehicles is the occupants.

[0102] In an embodiment of the present application, a damage risk assessment sub-model can be obtained in advance. After obtaining the first driving data and the second driving data, the computer device can input the first driving data and the second driving data into the damage risk assessment sub-model for evaluation to obtain the damage risk.

[0103] Specifically, such as Figure 6 As shown, the method for obtaining the above damage risk assessment sub-model includes:

[0104] S3031, obtaining collision speed sample data and other sample data of the vehicle under collision conditions.

[0105] Among them, the other sample data include at least one of initial collision condition sample data, occupant physiological parameter sample data and occupant status sample data.

[0106] In an embodiment of the present application, the computer may obtain collision speed sample data and other sample data of the vehicle under collision conditions by crawling the web or through a public traffic website.

[0107] S3032: Perform high-dimensional processing on the other sample data to obtain processed other sample data, and train an initial damage risk assessment sub-model based on the processed other sample data to obtain a damage risk assessment sub-model.

[0108] The initial damage risk assessment sub-model can be a neural network model or a machine learning model. The damage risk assessment sub-model can be a neural network model or a machine learning model.

[0109] In an embodiment of the present application, after the computer device obtains the collision speed sample data and other sample data of the vehicle under the collision condition based on the above steps, since the collision speed is a key quantitative indicator under the collision condition, in order to ensure the accuracy of the damage risk assessment sub-model, no high-dimensional processing is performed on the collision speed sample data, and the original value is retained. Only high-dimensional processing is performed on the other sample data to obtain the processed other sample data, and then the initial damage risk assessment sub-model is trained based on the processed other sample data to obtain the damage risk assessment sub-model.

[0110] For example, in a critical scenario, a computer can determine the approximate future collision conditions based on the current relative positions and velocities of the vehicles. It's worth noting that even in the event of a collision, vehicles with different levels of automation exhibit significant differences in occupant protection due to their avoidance strategies, optimized safety features, and varying driver habits. Key variables significantly impacting occupant injury outcomes can be summarized as: human, vehicle, and road. Specifically, these variables include: human factors—physiological characteristics of occupants (age, gender, height, weight, etc.) and seating posture (seat orientation, backrest angle, body position); vehicle factors—physical parameters (mass, size, stiffness, etc.) and restraint system information (seatbelts, airbags, etc.); and road factors—obstacle type (car, pedestrian, building), collision speed, and collision location. Consequently, occupant injury severity is rapidly and dynamically influenced by the interactive coupling of these multiple factors. The initial collision condition input and occupant injury output exhibit a complex, multi-layered mapping relationship, making it difficult to accurately establish a mathematical model.

[0111] In summary, a neural network-based occupant injury risk prediction model can be used to quantify the severity of occupant injuries in each potential collision or collision accident. The computer device can obtain the collision speed sample data and other sample data of the vehicle under the collision condition. The other sample data include at least one of the initial collision condition sample data, occupant physiological parameter sample data and occupant status sample data. Then, a neural network model based on a multilayer perceptron (MLP) can be used to directly predict the severity of occupant injuries in the collision based on the specific conditions at the moment of collision, occupant physiological parameters, restraint system parameters and occupant status data. Taking the severity of head injury as an example, the structure of the neural network model is as follows: Figure 7 The severity of other parts is calculated using the same algorithm:

[0112] The model consists of a data preprocessing module, a neural network processing module, and a result post-processing module:

[0113] Data preprocessing module (input layer): Since the initial crash conditions (crash velocity, impact angle, vehicle offset, and ground friction coefficient), occupant physiological parameters (gender, height, and weight), restraint system (seatbelts, head airbags, and knee airbags), and occupant posture (sitting position, posture, and distance from the steering wheel) are all scalar, the remaining scalar information, except for the crash velocity, is mapped into a high-dimensional space that is more suitable for neural network processing through the embedding layer, thereby enriching the input data features. Due to the significant quantitative relationship between crash velocity and occupant injuries (higher crash velocity tends to result in more severe injuries), the model uses crash velocity as a direct input without processing through the embedding layer.

[0114]

[0115]

[0116] in, is the scalar information other than the collision velocity, is the embedded high-dimensional feature vector, is the collision velocity, is the feature embedding function, is the feature concatenation function, is the input feature matrix.

[0117] Neural network processing module (hidden layer): The model consists of two fully connected layers and two normalized layers. The activation function can be ReLu function or tanh function.

[0118] a. Fully connected layer 1:

[0119]

[0120] b. Normalization layer 1:

[0121]

[0122] c. Fully connected layer 2:

[0123]

[0124] d. Normalization layer 2:

[0125]

[0126] in, and is the weight matrix and bias vector of the first fully connected layer, is the output of the first layer, and They are The mean and variance of and is the weight matrix and bias vector of the second fully connected layer, is the output of the second layer.

[0127] Result post-processing module (output layer): It consists of a linear layer, which maps the information processed by the neural network from high-dimensional space to low-dimensional space. After discretization, it finally outputs the occupant's head injury criterion (HIC).

[0128]

[0129]

[0130] in, and are the weight matrix and bias vector of the linear layer, is a low-dimensional feature vector. is the discretization function.

[0131] Model Training and Optimization: The database used for neural network model training can be generated through a joint simulation using finite element analysis software and multibody dynamics simulation software (such as MADYMO). First, the parameters of the specific collision scenario are determined. A high-precision vehicle model is constructed using finite element analysis software (such as LS-DYNA or ANSYS), and intermediate parameters such as the collision waveform are calculated. The collision waveform is then input into the multibody dynamics model (MADYMO), along with occupant and restraint system parameters. The resulting data is then calculated and output for each occupant's injury. This creates a database encompassing the specific collision scenarios and occupant injury severity.

[0132] The damage function can use the mean square error damage function or the cross entropy loss function. The model optimizer can use stochastic gradient descent (SGD, Adam, etc.). The mean square error loss function can be expressed as:

[0133]

[0134] in, is the actual occupant damage value (e.g. head HIC), is the predicted occupant damage value.

[0135] The above neural network can quantify the severity of injuries to various parts of the body. This can be represented by the Abbreviated Injury Scale (AIS), and the occupant injury risk in critical scenarios can be obtained by weighted summation:

[0136]

[0137] in, Indicates the weight of each body part, for example , , .

[0138] The methods described in the embodiments of this application not only perform data collection, model building, and preliminary verification in a driving simulator, but also design a migration strategy from the simulated environment to real-world road traffic scenarios. The model design inputs take into account the feasibility of real-world vehicle deployment. Multidimensional vehicle dynamics data readily available during daily operation, as well as operating condition characteristics during near-misses and collisions, are used as inputs to the safety benefit assessment model. During model development and optimization, natural driving databases and real-world accident databases are incorporated to enhance the model's accuracy and reliability. By analyzing the massive amount of driving behavior data in the natural driving database, specific driving behaviors and thresholds of actual drivers in different road traffic scenarios can be extracted. This allows the thresholds of the model weight parameters to be constrained in specific road traffic scenarios, providing model parameter settings that more closely reflect real-world driving behavior. Furthermore, by combining statistical data, accident characteristics, and the resulting economic losses (including property damage and medical insurance losses resulting from casualties) from the real-world accident database, the overall economic loss is used as an objective quantification of the unified safety benefit in the model. This allows the weights for driving risk, collision risk, and occupant injury risk to be adjusted to better align with industry statistics and reports on autonomous vehicles. This real-data-based fine-tuning process not only enhances the accuracy of the model's unified safety benefit assessment based on driving risk, collision risk, and occupant injury risk for vehicles at different levels of autonomy, but also verifies the model's performance under imperfect inputs by leveraging anomalies (noise and missing data) in real-world data, ensuring its robustness. After successful model validation, the model is deployed in a large fleet and traffic flow for real-world testing. During this process, real-time driving behavior and critical condition data are collected through onboard systems or roadside connected devices to monitor the unified safety benefit output of the model. This ensures the feasibility and universality of the technical solution in real-world road scenarios, enabling real-time, long-term safety benefit assessments for vehicles at different levels of autonomy to improve road safety. The collection of driving behavior and collision condition data for occupant injury risk prediction in critical conditions is achieved. Using a high-fidelity driving simulator and a multi-dimensional data acquisition system, comprehensive vehicle, scenario, and driver-level data is collected across various scenarios. A neural network model is then used to rapidly quantify the severity of occupant injuries in potential collision accidents under hazardous scenarios. This ensures the completeness of data under hazardous scenarios and the reliability of the input data for the safety-benefit assessment model. A real-time safety-benefit assessment model for hazardous scenarios, combining driving risk, collision risk, and occupant injury risk, has been constructed. This invention develops a unified safety-benefit assessment model applicable to vehicles of varying levels of autonomous driving, covering normal driving scenarios, hazardous scenarios, and collision accident scenarios.It also proposes a technical path from data verification based on driving simulators, to fine-tuning of model parameters based on natural driving data and real accident data, and finally to large-scale real-vehicle application, to ensure that the model can adapt to various road traffic scenarios and accurately and objectively quantify the safety benefits of cars of different autonomous driving levels, providing an objective safety basis for the development of autonomous driving cars.

[0139] In some embodiments, the method for evaluating the safety of an autonomous vehicle during its entire driving process includes:

[0140] S401, obtaining a safety assessment scale.

[0141] The safety assessment scale includes multiple driving factors, each driving factor includes multiple driving safety questions, and scores corresponding to each driving safety question.

[0142] In the embodiment of the present application, a large number of safety assessment scales can be obtained by volunteers participating in simulation experiments, or based on the autonomous driving car acceptance scale commonly used in existing studies, the safety-related content is screened, and the "perceived safety" part of the scale and the "experimental validity" part used to determine the validity of the experiment and the scale are supplemented with the characteristics of the dangerous scene in the experiment to form a large number of self-safety assessment scales. The safety assessment scale contains 43 items in total. Participants are asked to read each question and score it according to the degree of fit between the question and their feelings, thoughts and behaviors when driving the corresponding autonomous driving level car. The scoring adopts a 5-point scale ranging from "absolutely disagree (1)" to "absolutely agree (5)". Six questions are used to determine whether the scale is valid (Q38-Q40: verify the consistency between the behavior in the experiment and the real driving; Q41-Q43: verify the authenticity and closeness of the experimental scene). The 36 questions are divided into 6 factors. Each factor represents a specific feature description of the subjective quantification of the safety of autonomous driving cars. According to the scores of different factors, the subjective safety assessment of cars of different autonomous driving levels can be comprehensively quantified. The specific safety assessment scale is shown in the following table:

[0143]

[0144] S402: Construct a safety assessment model based on the safety assessment scale and the driving risk parameters, collision risk parameters, and injury risk parameters.

[0145] In an embodiment of the present application, after the computer device obtains the safety assessment scale based on the above steps, it can construct a safety assessment model based on the safety assessment scale and driving risk parameters, collision risk parameters and injury risk parameters.

[0146] Specifically, the step of "building a safety assessment model based on the safety assessment scale and the driving risk parameters, the collision risk parameters, and the injury risk parameters" in S402 includes:

[0147] S4031, construct a driving factor loading matrix based on the safety assessment scale.

[0148] The driving factor loading matrix represents the correlation between each question (positive correlation, negative correlation, or no correlation).

[0149] S4032: Determine the load value corresponding to each driving factor according to the driving factor load matrix.

[0150] S4033: Determine the weight of the driving risk parameter, the weight of the collision risk parameter, and the weight of the injury risk parameter based on the load value corresponding to each driving factor and the score corresponding to each driving safety question.

[0151] In the embodiment of the present application, after obtaining the safety assessment scale (hereinafter referred to as the scale), the computer device first evaluates the effectiveness of the experiment for the volunteer by scoring according to the "experimental effectiveness" (i.e., questions 38-43) in the safety assessment scale. If the score is higher than the preset value (e.g., 80 points), it means that the safety assessment scale is valid, and the safety assessment scale can be used to continue the analysis. If it is invalid, the safety assessment scale is deleted and the validity of other safety assessment scales is reconfirmed. Secondly, other parts (overall trust, driving / riding emotion, perceived safety, perceived usefulness, perceived ease of use) can be used to verify whether the behavior of the volunteers in the experiment is consistent with the dimensions measured by the scale. Specifically, this can be done by obtaining the driving behavior data of the volunteers during the driving process (such as accelerator pedal force ( ), brake pedal force ( )、Steering wheel angle( )、Steering wheel torque( )) and the overall trust, driving / riding emotions, and perceived ease of use in the scale are analyzed for correlation (for example, verify the correlation based on whether emergency braking, large-scale steering, and other behaviors correspond to driving emotions). For example, the brake pedal force recorded in the dangerous scenario in the experiment ( ) can be directly compared with the scores of worry or anxiety questions in the driving / riding emotion questions in the scale. For example, when the volunteers are driving, they always generate brake pedal force, which means that the emotion during driving / riding is worry or anxiety. Then we can check whether the generation of brake pedal force is positively correlated with worry or anxiety. Finally, after the scale statistics are completed, the question scores are verified to ensure that the measured variables conform to the expected psychological or statistical logic. For example, relaxed driving emotions and trust in autonomous driving technology should be positively correlated (for example, the linear relationship of the variables can be verified by calculating the Pearson correlation technique).

[0152] In order to quantify the impact weights of driving risk, collision risk, and injury risk on unified safety benefits, the following method is used:

[0153] ① The score data of each question in the scale is normalized to a mean of 0 and a variance of 1 to eliminate the differences in the scoring scales of different questions and make the scores of different subjects comparable.

[0154]

[0155] in, Indicates the score of the test-taker on this question. represents the mean value of each question, represents the standard deviation of each question.

[0156] ② Factor analysis, calculate the factor loading of each question

[0157] a. First calculate the correlation matrix R (i.e., driving factor loading matrix) of the scale data to indicate the correlation between each question (positive correlation, negative correlation, or no correlation):

[0158]

[0159] in:

[0160]

[0161]

[0162]

[0163] For each pair of questions (i, j), calculate the Pearson correlation coefficient:

[0164]

[0165] in, is the standardized score of question i; is the standardized score of question j; n is the number of subjects or the number of safety assessment scales.

[0166] b. Perform principal component analysis (PCA): Calculate the eigenvalues and select factors with eigenvalues greater than 1 according to the Kaiser criterion to obtain the loading matrix F. Then perform Varimax rotation to obtain the final driving factor loading matrix. (After rotation, the factor loadings are clearer), and from Extract the loading value on the corresponding factor ,in, include 、 、 … 、 、 , It can be expressed as m×n, where m represents the question and n represents the factor (i.e. the first column "factor" in the safety assessment table). The driving factor loading matrix for:

[0167]

[0168] ③Calculate the impact weight of each risk (driving, collision, injury) on the unified safety benefit

[0169] According to the scale content, select questions related to driving, collision, and injury risk from the questions, and combine them with the scale scores Quantify the weights of driving, collision, and injury risks:

[0170] The classification standards for topics can refer to the following standards:

[0171] Driving risk: 1-2, 4-7, 15-16, 19, 21-22, 24, 29;

[0172] Collision risk: 5-7, 14, 18, 20, 23, 27-33;

[0173] Injury risk: 3-7, 16-18, 25-28, 30;

[0174]

[0175]

[0176]

[0177] in, Represents the weight of the driving risk parameter; represents the weight of the collision risk parameter; Represents the weight of the damage risk parameter.

[0178] S5034: Construct a safety assessment model based on the weights of the driving risk parameters, the weights of the collision risk parameters, the weights of the injury risk parameters, the driving risk parameters, the collision risk parameters, and the injury risk parameters.

[0179] In the embodiment of the present application, after the computer device obtains the weights of the driving risk parameters, the weights of the collision risk parameters, and the weights of the injury risk parameters based on the above steps, it can construct a safety assessment model based on the weights of the driving risk parameters, the weights of the collision risk parameters, the weights of the injury risk parameters, the driving risk parameters, the collision risk parameters, and the injury risk parameters. The specific safety assessment model is as follows:

[0180]

[0181] in, represents the safety assessment value; Indicates driving risk; Indicates collision risk; Indicates risk of injury; represents the driving risk parameter; represents the collision risk parameter; represents the injury risk parameter; Indicates the total mileage; Indicates the total driving time within the preset area.

[0182] The method described in the embodiments of the present application can effectively collect data on the entire process of autonomous vehicles in dangerous scenarios, including driving behavior, collision conditions, and occupant injury risks, and propose a unified real-time safety benefit evaluation model that integrates driving risk, collision risk, and occupant injury risk, thereby objectively and reasonably evaluating the safety of future road traffic scenarios.

[0183] In summary of all the above embodiments, a method for safety assessment of the entire driving process of an autonomous vehicle is also provided, which includes:

[0184] S601: Obtain a safety assessment scale, construct a driving factor load matrix based on the safety assessment scale, determine the load value corresponding to each driving factor based on the driving factor load matrix, determine the weight of the driving risk parameter, the weight of the collision risk parameter, and the weight of the injury risk parameter based on the load value corresponding to each driving factor and the score corresponding to each driving safety question, and construct a safety assessment model based on the driving risk parameter weight, the collision risk parameter weight, the injury risk parameter weight, the driving risk parameter, the collision risk parameter, and the injury risk parameter. The safety assessment scale includes multiple driving factors, each driving factor includes multiple driving safety questions, and the scores corresponding to each driving safety question.

[0185] S602: Obtain collision velocity sample data and other sample data of the vehicle under a collision condition, perform high-dimensional processing on the other sample data to obtain processed other sample data, and train an initial damage risk assessment sub-model based on the processed other sample data to obtain a damage risk assessment sub-model. The other sample data includes at least one of initial collision condition sample data, occupant physiological parameter sample data, and occupant posture sample data.

[0186] S603: Acquire first driving data of the target vehicle, and acquire second driving data of vehicles in an area surrounding the target vehicle.

[0187] At step S604, the first driving data is input into the weight analysis sub-model to obtain weights of target factors, which are then input into the first calculation sub-model to obtain driving risk. The target factors include vehicle speed, acceleration, lane departure distance, and environmental factors.

[0188] S605: Input the first driving data and the second driving data into the analysis sub-model to obtain the longitudinal collision risk and the lateral collision risk, and input the longitudinal collision risk and the lateral collision risk into the second calculation sub-model to obtain the collision risk.

[0189] S606: Input the first driving data and the second driving data into a damage risk assessment sub-model for assessment to obtain damage risk.

[0190] S607: Input the driving risk, collision risk, and damage risk into the safety assessment model to perform a safety assessment on the target vehicle and obtain a safety assessment result.

[0191] In the embodiments of this application, vehicle driving risk refers to the fact that each component of the traffic system has the potential to create driving risk. The risk posed to vehicles on the road is objective and not subject to human subjective will. During vehicle operation, its safety is affected by numerous factors, and these factors (such as driver behavior, vehicle driving conditions, and road traffic environment) change dynamically. Therefore, the likelihood, degree, and type of driving risk also change dynamically and are uncertain. The driving risk of an autonomous vehicle reflects its safety and reliability in the complex and ever-changing real-world traffic environment. Vehicle collision risk and accident conditions: Vehicle collision risk refers to the probability and potential severity of a physical collision with another vehicle, pedestrian, object, or obstacle during dynamic driving, due to the influence of the external environment, vehicle performance, or driving behavior. Collision risk assessment is an important indicator for measuring vehicle safety performance, helping to predict and reduce accidents and improve driving safety. Accident conditions describe the environmental conditions and dynamic response of a vehicle during a collision, as well as related physical parameters. Specifically, these parameters include vehicle-level (collision speed, angle, location, overlap ratio), road-level (road friction coefficient, nature of the impacting object), occupant-level (height, weight), and restraint system-level (seatbelt usage, seatbelt force limiter, airbag deployment). These specific operating conditions can be used to analyze the causes of collisions and changes in vehicle dynamics, further quantifying the severity of the collision and the potential injury risk to occupants, providing a critical basis for accident reconstruction and safety assessment. Regarding occupant injury risk, for traffic participants, injury refers to the deformation of human anatomical structures under external loads exceeding their failure limits, leading to tissue damage or structural failure. The damage mechanisms and failure limits of key body parts are crucial data references for vehicle safety assessment and traffic participant protection. At the moment of a collision, the initial kinetic energy carried by the vehicle during normal driving is rapidly released and transferred to the vehicle body and occupants, causing deformation or damage, resulting in damage to the vehicle body and the risk of occupant injury. The type and severity of injuries sustained by traffic participants are difficult to directly assess. Instead, they must be converted into injury indicators corresponding to specific body parts using intuitively measurable physical parameters such as acceleration, velocity, displacement, compression or extension, force, and stress. Commonly used evaluation indicators include the Head Injury Criterion and maximum chest deflection.

[0192] The methods described in the examples of this application are based on existing technologies that lack full-process data covering normal driving, critical scenarios, and collision accidents. In autonomous vehicle safety research, evaluating the safety benefits of autonomous vehicles requires multi-dimensional, full-process data support. However, existing data typically relies on natural driving databases lacking critical scenarios and real-world accident databases with limited data and incomplete occupant injury information. Fragmented data collection methods and single-source data with low information completeness fail to fully utilize all key safety-related information in road traffic scenarios across all time dimensions, making it difficult to provide data support for accurately quantifying the safety benefits of autonomous vehicles. Existing technologies lack a unified safety benefit assessment model that integrates driving, collision, and occupant injury risks. Existing safety benefit assessment criteria for autonomous vehicles are relatively simplistic, mostly quantifying only driving and collision risks, ignoring the risk of occupant injury in collision accidents in critical scenarios. Occupants are the primary protected population of smart vehicles and a key factor influencing autonomous vehicle safety. Therefore, this application comprehensively considers driving, collision, and occupant injury risks in critical scenarios based on driving behavior, facilitating real-time and objective assessment of the safety benefits of autonomous vehicles.

[0193] This application proposes a data collection system for the entire "driving-collision-damage" process in critical scenarios. This technology develops control algorithms for vehicles of different autonomous driving levels by designing normal and critical scenarios with varying levels of traffic risk. Driver experiments and full-process data collection are conducted using a driving simulator. This multi-scenario, multi-autonomous driving level and data collection system is an integrated solution designed to comprehensively collect and analyze driver behavior data when operating vehicles of different autonomous driving levels in various traffic scenarios, as well as collision condition data in critical scenarios, using a high-fidelity driving simulator. The core of the system consists of a multi-degree-of-freedom, high-fidelity driving simulator equipped with advanced dynamic feedback mechanisms and multi-dimensional data collection technology, integrated with control algorithms adapted for vehicles of different autonomous driving levels. Designed with a focus on a realistic driving experience, this driving simulator supports comprehensive dynamics simulation, including vehicle physical responses such as acceleration, braking, steering, and road conditions. The simulator also provides multi-sensory feedback, including visual, auditory, and tactile feedback, that closely resembles the actual driving experience, ensuring a highly realistic driving experience in the virtual environment. The system's data collection is extensive and comprehensive, covering multiple dimensions. Specifically, it includes: vehicle information collection (kinematic and dynamic data of the vehicle during normal driving, dangerous scenarios, and at the moment of collision, dynamic information of surrounding vehicles, and data captured by on-board cameras), road section information collection (the urgency of traffic scenarios, the type of dangerous scenarios, road environment characteristics, and ground friction coefficient, etc.), and driver information collection (driving behavior, sitting posture, and restraint system status are collected through motion capture and depth imaging technology. At the same time, the survey is combined with identity information such as age and driving experience.

[0194] This application develops a real-time safety benefit evaluation model for autonomous vehicles. The present invention innovatively combines the driving risk and collision risk during normal driving in dangerous scenarios, and quantifies the injury risk of occupants in potential collision accidents through an occupant injury risk prediction model, for real-time and comprehensive evaluation of the safety benefit of autonomous vehicles in dangerous scenarios. This application considers model verification and application in real-world road traffic environments. The present invention uses normal driving data based on a driving simulator, driving behavior data in dangerous scenarios, collision condition data, and an occupant injury risk prediction model to establish a real-time safety benefit evaluation model for autonomous vehicles in dangerous scenarios, and verifies the effectiveness of the model, and finally proposes a migration strategy from a simulated environment to a real road traffic scenario. This application is based on a high-fidelity driving simulator, and this technical solution can be applied to cars of different autonomous driving levels, simulating normal traffic scenarios and multiple types of dangerous traffic scenarios, and collecting the behavioral characteristics of drivers when driving cars of different autonomous driving levels. This paper utilizes an occupant injury risk prediction model, combined with driver behavior and vehicle dynamics data from both normal and critical traffic scenarios, to develop a novel safety benefit assessment model for autonomous vehicles in critical scenarios. This model extends safety assessment from critical scenarios forward to normal driving and backward to collisions and occupant injury, extending safety benefit assessment from traditional driving risk quantification to collision risk quantification and occupant injury risk assessment. It also explores the relationship between the safety benefit and driving behavior of autonomous vehicles in critical scenarios. This paper primarily comprises three components: a full-process data acquisition system for critical scenarios, a unified safety benefit assessment model for autonomous vehicles, and model validation and application considering real-world road conditions.

[0195] It should be understood that, although the various steps in the flowcharts involved in the various embodiments described above are displayed in sequence according to the instructions of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the various embodiments described above can include multiple steps or multiple stages, and these steps or stages are not necessarily executed and completed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0196] Based on the same inventive concept, embodiments of the present application also provide an autonomous vehicle full-process driving safety assessment device for implementing the aforementioned autonomous vehicle full-process driving safety assessment method. The solution provided by this device is similar to the solution described in the aforementioned method. Therefore, the specific limitations of one or more embodiments of the autonomous vehicle full-process driving safety assessment device provided below can be found in the aforementioned limitations of the autonomous vehicle full-process driving safety assessment method and will not be further elaborated here.

[0197] In some embodiments, as Figure 8 As shown, a device for evaluating the safety of an autonomous vehicle during its entire driving process is provided, comprising:

[0198] The acquisition module is used to acquire first driving data of a target vehicle and second driving data of vehicles in an area surrounding the target vehicle.

[0199] The risk assessment module is used to input the first driving data and the second driving data into the risk assessment model to perform risk assessment on the target vehicle and obtain a risk assessment result; the risk assessment result includes driving risk, collision risk and damage risk.

[0200] The safety assessment module is used to input the risk assessment results into the safety assessment model to perform a safety assessment on the target vehicle and obtain a safety assessment result.

[0201] In some embodiments, the risk assessment module includes:

[0202] The first evaluation unit is configured to input the first driving data into the driving risk evaluation sub-model for evaluation to obtain the driving risk.

[0203] The second evaluation unit is configured to input the first driving data and the second driving data into the collision risk evaluation sub-model for evaluation to obtain the collision risk.

[0204] The third evaluation unit is configured to input the first driving data and the second driving data into the damage risk evaluation sub-model for evaluation to obtain the damage risk.

[0205] In some embodiments, the above-mentioned first evaluation unit is specifically used to input the first driving data into the weight analysis sub-model to obtain the weight of the target factor; the target factor includes vehicle speed factor, acceleration factor, lane departure distance factor and environmental factor; the weight of the target factor is input into the first calculation sub-model to obtain the driving risk.

[0206] In some embodiments, the above-mentioned second evaluation unit is specifically used to input the first driving data and the second driving data into the analysis sub-model to obtain the longitudinal collision risk and the lateral collision risk; and input the longitudinal collision risk and the lateral collision risk into the second calculation sub-model to obtain the collision risk.

[0207] In some embodiments, the above-mentioned risk assessment module is also used to obtain collision speed sample data and other sample data of the vehicle under collision conditions; the other sample data include at least one of initial collision condition sample data, occupant physiological parameter sample data and occupant posture sample data; high-dimensional processing is performed on the other sample data to obtain processed other sample data, and the initial damage risk assessment sub-model is trained based on the processed other sample data to obtain a damage risk assessment sub-model.

[0208] In some embodiments, the above-mentioned autonomous driving vehicle driving process safety assessment device is also used to obtain a safety assessment scale; the safety assessment scale includes multiple driving factors, each driving factor includes multiple driving safety questions, and the scores corresponding to each driving safety question; a safety assessment model is constructed based on the safety assessment scale and driving risk parameters, collision risk parameters and injury risk parameters.

[0209] In some embodiments, the above-mentioned autonomous driving vehicle driving whole-process safety assessment device is further used to construct a driving factor load matrix based on a safety assessment scale; determine the load value corresponding to each driving factor based on the driving factor load matrix; determine the weight of the driving risk parameter, the weight of the collision risk parameter, and the weight of the injury risk parameter based on the load value corresponding to each driving factor and the score corresponding to each driving safety question; and construct a safety assessment model based on the weight of the driving risk parameter, the weight of the collision risk parameter, the weight of the injury risk parameter, the driving risk parameter, the collision risk parameter, and the injury risk parameter.

[0210] Each module in the aforementioned autonomous vehicle full-process safety assessment device can be implemented in whole or in part through software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor within a computer device in hardware form, or stored in a computer device's memory in software form, allowing the processor to call and execute the corresponding operations of each module.

[0211] In some embodiments, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the steps of the method for safety assessment of the entire driving process of an autonomous driving vehicle described in any of the above embodiments are implemented.

[0212] In some embodiments, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of the method for evaluating the safety of the entire driving process of an autonomous driving vehicle described in any of the above embodiments are implemented.

[0213] In some embodiments, a computer program product is provided, including a computer program, which, when executed by a processor, implements the steps of the method for assessing the safety of the entire driving process of an autonomous driving vehicle as described in any of the above embodiments.

[0214] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the above-mentioned embodiments. In particular, any reference to memory, database, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The databases involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processors involved in the various embodiments provided herein may be, but are not limited to, general-purpose processors, central processing units (CPUs), graphics processing units (GPUs), digital signal processors (DSPs), programmable logic devices (PLDs), data processing logic devices based on quantum computing, and the like.

[0215] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0216] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present application. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present application, and these modifications and improvements fall within the scope of protection of the present application. Therefore, the scope of protection of the present application shall be determined by the appended claims.

Claims

1. A method for evaluating the safety of an autonomous vehicle during its entire driving process, characterized in that: The method comprises: Acquiring first driving data of a target vehicle, and acquiring second driving data of vehicles in an area surrounding the target vehicle; Inputting the first driving data and the second driving data into a risk assessment model to perform risk assessment on the target vehicle to obtain a risk assessment result; the risk assessment result includes driving risk, collision risk, and damage risk; The risk assessment result is input into a safety assessment model to perform a safety assessment on the target vehicle to obtain a safety assessment result.

2. The method according to claim 1, characterized in that The risk assessment model includes a driving risk assessment sub-model, a collision risk assessment sub-model, and a damage risk assessment sub-model. The first driving data and the second driving data are input into the risk assessment model to perform risk assessment on the target vehicle, and a risk assessment result is obtained, including: Inputting the first driving data into the driving risk assessment sub-model for assessment to obtain the driving risk; Inputting the first driving data and the second driving data into the collision risk assessment sub-model for assessment to obtain the collision risk; The first driving data and the second driving data are input into the damage risk assessment sub-model for assessment to obtain the damage risk.

3. The method according to claim 2, characterized in that The driving risk assessment sub-model includes a weight analysis sub-model and a first calculation sub-model. Inputting the first driving data into the driving risk assessment sub-model for assessment to obtain the driving risk includes: Inputting the first driving data into the weight analysis sub-model to obtain weights of target factors; the target factors include vehicle speed factor, acceleration factor, lane departure distance factor and environmental factor; The weight of the target factor is input into the first calculation sub-model to obtain the driving risk.

4. The method according to claim 2, characterized in that The collision risk assessment sub-model includes an analysis sub-model and a second calculation sub-model. Inputting the first driving data and the second driving data into the collision risk assessment sub-model for assessment to obtain the collision risk includes: Inputting the first driving data and the second driving data into the analysis sub-model to obtain a longitudinal collision risk and a lateral collision risk; The longitudinal collision risk and the lateral collision risk are input into the second calculation sub-model to obtain the collision risk.

5. The method according to claim 2, characterized in that The method for obtaining the damage risk assessment sub-model includes: Acquiring collision velocity sample data and other sample data of the vehicle under a collision condition; the other sample data includes at least one of initial collision condition sample data, occupant physiological parameter sample data, and occupant posture sample data; High-dimensional processing is performed on the other sample data to obtain processed other sample data, and an initial damage risk assessment sub-model is trained based on the processed other sample data to obtain the damage risk assessment sub-model.

6. The method according to any one of claims 1 to 5, characterized in that The method further comprises: Obtaining a safety assessment scale; the safety assessment scale includes multiple driving factors, each of the driving factors includes multiple driving safety questions, and scores corresponding to each driving safety question; The safety assessment model is constructed based on the safety assessment scale and the driving risk parameters, collision risk parameters and injury risk parameters.

7. The method according to claim 6, characterized in that The step of constructing the safety assessment model according to the safety assessment scale and the driving risk parameter, the collision risk parameter, and the injury risk parameter includes: Constructing a driving factor loading matrix based on the safety assessment scale; Determining a load value corresponding to each driving factor according to the driving factor load matrix; Determining the weight of the driving risk parameter, the weight of the collision risk parameter, and the weight of the injury risk parameter according to the load value corresponding to each driving factor and the score corresponding to each driving safety question; The safety assessment model is constructed according to the weight of the driving risk parameter, the weight of the collision risk parameter, the weight of the injury risk parameter, the driving risk parameter, the collision risk parameter and the injury risk parameter.

8. A device for evaluating the safety of an autonomous vehicle during its entire driving process, characterized in that: The device comprises: an acquisition module, configured to acquire first driving data of a target vehicle and second driving data of vehicles in an area surrounding the target vehicle; a risk assessment module, configured to input the first driving data and the second driving data into a risk assessment model to perform a risk assessment on the target vehicle and obtain a risk assessment result; the risk assessment result includes driving risk, collision risk, and damage risk; The safety assessment module is used to input the risk assessment result into the safety assessment model to perform a safety assessment on the target vehicle and obtain a safety assessment result.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images