Complexity evaluation method and device for automatic driving reference scene library, equipment and medium

Abstract

The invention discloses an automatic driving reference scene library complexity evaluation method and device, equipment and a medium, and relates to the technical field of intelligent driving, and the method comprises the steps: obtaining to-be-evaluated automatic driving scene data; screening out elements influencing the complexity of the automatic driving scene according to a preset standard rule; based on an analytic hierarchy process, constructing a judgment matrix of the elements, and calculating a weight coefficient of each element according to the judgment matrix; according to the types of the elements, the independent complexity of each element is calculated, and the types of the elements comprise Boolean value types and non-Boolean value types; calculating the complexity of the automatic driving scene at any moment according to the independent complexity of the elements and the weight coefficient; calculating the average complexity of the automatic driving scene according to the complexity of the automatic driving scene at the multiple moments; and according to the average complexity, grading the automatic driving scene to obtain a complexity grade of the automatic driving scene.

Description

Technical Field

[0001] The present application relates to the field of intelligent driving technology, and in particular to a method, device, equipment and medium for evaluating the complexity of an autonomous driving benchmark scenario library. Background Art

[0002] With the rapid development of autonomous driving technology, autonomous vehicles need to operate safely and reliably in complex and changing traffic environments. In order to ensure the performance of autonomous driving systems in different scenarios, their environmental perception, decision-making planning and execution capabilities must be comprehensively evaluated. Scenario complexity evaluation is a key link in the testing and verification of autonomous driving systems. It can quantify the impact of different scenarios on autonomous driving systems, help R&D personnel identify system performance bottlenecks, optimize algorithm design, and improve system safety and reliability.

[0003] Related technologies usually use preset rules and thresholds to simply classify and score the elements in the scene. For example, the scene complexity is roughly divided according to static features such as the number of traffic participants and road types. This method is simple and easy to implement, but lacks quantitative analysis of dynamic elements (such as weather changes and traffic flow dynamics), and it is difficult to accurately reflect the real-time complexity of the scene.

[0004] Alternatively, we can use historical data to build a statistical model and evaluate the complexity of the scene by analyzing the distribution and correlation of various elements in the scene. This method can reflect the complexity of the scene to a certain extent, but it is highly dependent on data and has difficulty in handling nonlinear relationships and dynamic changes.

[0005] Alternatively, we can train machine learning models to automatically extract features from scenes and predict complexity. This method can handle complex nonlinear relationships, but it requires a large amount of labeled data, and the model is less interpretable, making it difficult to clearly identify the specific contribution of each factor to complexity. Summary of the invention

[0006] The purpose of this application is to provide a method, device, equipment and medium for evaluating the complexity of an autonomous driving benchmark scenario library.

[0007] To achieve the above objectives, this application provides the following solutions:

[0008] In a first aspect, the present application provides a method for evaluating the complexity of an autonomous driving benchmark scenario library, comprising:

[0009] Obtain the autonomous driving scenario data to be evaluated;

[0010] Screening out factors that affect the complexity of the autonomous driving scenario according to preset standard regulations;

[0011] Based on the hierarchical analysis method, a judgment matrix of the elements is constructed, and a weight coefficient of each element is calculated according to the judgment matrix;

[0012] According to the type of the element, the independent complexity of each element is calculated respectively, and the type of the element includes a Boolean value type and a non-Boolean value type;

[0013] Calculating the complexity of the autonomous driving scenario at any moment according to the independent complexity of the elements and the weight coefficients;

[0014] Calculating an average complexity of the autonomous driving scenario according to the complexity of the autonomous driving scenario at multiple moments;

[0015] The autonomous driving scenarios are graded according to the average complexity to obtain complexity levels of the autonomous driving scenarios.

[0016] In a second aspect, the present application provides a device for evaluating the complexity of an autonomous driving benchmark scenario library, comprising:

[0017] An acquisition module, used to acquire the autonomous driving scene data to be evaluated;

[0018] An evaluation module, used to screen out factors that affect the complexity of the autonomous driving scenario according to preset standard regulations;

[0019] Based on the hierarchical analysis method, a judgment matrix of the elements is constructed, and a weight coefficient of each element is calculated according to the judgment matrix;

[0020] According to the type of the element, the independent complexity of each element is calculated respectively, and the type of the element includes a Boolean value type and a non-Boolean value type;

[0021] Calculating the complexity of the autonomous driving scenario at any moment according to the independent complexity of the elements and the weight coefficients;

[0022] Calculating an average complexity of the autonomous driving scenario according to the complexity of the autonomous driving scenario at multiple moments;

[0023] The autonomous driving scenarios are graded according to the average complexity to obtain complexity levels of the autonomous driving scenarios.

[0024] In a third aspect, the present application provides a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of any one of the above-mentioned methods for evaluating the complexity of an autonomous driving benchmark scenario library.

[0025] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the above-mentioned methods for evaluating the complexity of an autonomous driving benchmark scenario library.

[0026] In a fifth aspect, the present application provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of any of the above-mentioned methods for evaluating the complexity of the autonomous driving benchmark scenario library.

[0027] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0028] The present application provides a method, device, equipment and medium for evaluating the complexity of an autonomous driving benchmark scenario library. By obtaining multi-dimensional autonomous driving scenario data and screening complexity factors in combination with the six-level principle in the preset standard, the comprehensiveness and standardization of the evaluation are ensured. The judgment matrix is ​​constructed and the weight coefficient is calculated using the hierarchical analysis method, and the influence of each factor on the complexity of the scenario is refined and quantified, avoiding the subjectivity of weight allocation in traditional methods. By calculating the independent complexity of Boolean and non-Boolean elements respectively, the actual influence of different elements is dynamically reflected, especially the complexity calculation of non-Boolean elements, which can more accurately reflect the real-time changes of the scenario. Furthermore, by calculating the complexity of the scenario at any moment and combining the complexity of multiple moments to calculate the average complexity, the complexity of the autonomous driving scenario can be reflected in real time, providing data support for system decision-making. Finally, the scenario is graded according to the average complexity, and the complexity level is intuitively divided, which is convenient for testers to understand and apply. The method has strong scalability, can flexibly adjust the factor screening and weight calculation according to needs, and is suitable for the evaluation of various complex scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0030] Figure 1 A schematic diagram of a process flow of a complexity evaluation method for an autonomous driving benchmark scenario library provided in one embodiment of the present application;

[0031] Figure 2 A schematic diagram of the structure of a scene element influence transfer model provided in one embodiment of the present application;

[0032] Figure 3A logical schematic diagram of a complexity evaluation method for an autonomous driving benchmark scenario library provided in one embodiment of the present application;

[0033] Figure 4 A schematic diagram of functional modules of a device for evaluating the complexity of an autonomous driving benchmark scenario library provided in one embodiment of the present application;

[0034] Figure 5 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0035] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0036] like Figure 1 As shown, some embodiments of the present application provide a method for evaluating the complexity of an autonomous driving benchmark scenario library, which includes the following steps 101 to 107. Among them:

[0037] Step 101: Obtain autonomous driving scene data to be evaluated.

[0038] In the embodiment of the present application, the system collects and obtains autonomous driving scene data for evaluation. These data usually come from sensors of autonomous driving vehicles, such as LiDAR, cameras, millimeter wave radar, ultrasonic sensors, and GPS. These sensors can perceive the surrounding environment in real time and accurately, including road conditions, traffic signs, pedestrians, other vehicles and other information, and collect a large amount of raw data. These data include the vehicle's position, speed, acceleration, yaw angle information, as well as the position and speed of surrounding vehicles and pedestrians.

[0039] Step 102: Filter out factors that affect the complexity of the autonomous driving scenario according to preset standards.

[0040] In the embodiment of the present application, the key factors affecting the complexity of the autonomous driving scene are screened out from these data. These factors are usually divided into multiple levels, such as road grade, traffic facilities, temporary traffic changes, traffic participants, weather conditions, vehicle status, etc. According to the six-level principle in ISO 34502, these factors can be further refined into specific scene complexity factors, such as the number of lanes, slope, turning curvature, signal lights, traffic signs, warning signs, vehicle speed, target vehicle position and speed, weather conditions (rain, snow, fog, lighting, etc.), etc.

[0041] Step 103: construct a judgment matrix of the elements based on the hierarchical analysis method, and calculate the weight coefficient of each element according to the judgment matrix.

[0042] In the embodiment of the present application, the analytic hierarchy process (AHP) is a commonly used and effective means to solve complex evaluation problems. In this step, it is first necessary to build a hierarchical model based on the intrinsic relationship between the elements. Then, based on expert experience or statistical data, the elements are compared in pairs to construct a judgment matrix. The elements in the judgment matrix represent the relative importance of each element. By calculating the maximum eigenvalue of the judgment matrix and its corresponding eigenvector, the eigenvector can be normalized to obtain the weight coefficient of each element. These weight coefficients reflect the degree of influence of each element on the complexity of the autonomous driving scenario.

[0043] Step 104, calculating the independent complexity of each of the elements according to the type of the elements, the type of the elements including Boolean type and non-Boolean type.

[0044] In an embodiment of the present application, the independent complexity of the screened scene complexity elements is calculated according to their type (Boolean type or non-Boolean type). For Boolean type elements (such as whether a traffic light exists, whether there is rain, etc.), if they appear in the scene, their own complexity is 1, otherwise it is 0. For non-Boolean type elements (such as light intensity, number of lanes, slope, etc.), it is necessary to substitute their specific values ​​into the corresponding complexity formula for calculation to obtain their independent complexity. These complexity formulas are usually constructed based on experience or statistical data to reflect the impact of each element on the complexity of the scene.

[0045] Step 105: Calculate the complexity of the autonomous driving scenario at any moment based on the independent complexity of the elements and the weight coefficients.

[0046] In the embodiment of the present application, after obtaining the independent complexity and weight coefficient of each element, the next step is to calculate the complexity of the autonomous driving scene at any time based on these parameters. This is usually achieved by weighted summation, that is, multiplying the independent complexity of each element by its corresponding weight coefficient, and then adding all the products to obtain the scene complexity at that moment.

[0047] Step 106: Calculate the average complexity of the autonomous driving scenario based on the complexity of the autonomous driving scenario at multiple moments.

[0048] In the embodiment of the present application, since the autonomous driving scene is dynamically changing, its complexity needs to be evaluated at multiple times. After obtaining the scene complexity at each moment, the next step is to calculate the average of these complexities to reflect the average complexity of the entire autonomous driving scene. This helps to understand the complexity of the scene more comprehensively and provides a scientific basis for subsequent testing and verification.

[0049] Step 107: Classify the autonomous driving scenarios according to the average complexity to obtain the complexity level of the autonomous driving scenarios.

[0050] In the embodiment of the present application, this is achieved by comparing the average complexity with a preset complexity level standard. For example, the complexity level can be divided into six levels: low complexity, medium-low complexity, medium complexity, medium-high complexity, high complexity, and extremely high complexity. Each level corresponds to a complexity range, and by comparing the relationship between the average complexity and these ranges, the complexity level of the autonomous driving scene can be determined. This helps testers to understand the complexity of the scene more intuitively and provide targeted guidance for subsequent testing and verification.

[0051] The embodiment of the present application obtains multi-dimensional autonomous driving scene data and selects complexity factors in combination with the six-level principle in the preset standard regulations to ensure the comprehensiveness and standardization of the evaluation. The judgment matrix is ​​constructed and the weight coefficient is calculated using the hierarchical analysis method to finely quantify the impact of each factor on the scene complexity, avoiding the subjectivity of weight allocation in traditional methods. By calculating the independent complexity of Boolean and non-Boolean elements respectively, the actual impact of different elements is dynamically reflected, especially the complexity calculation of non-Boolean elements, which can more accurately reflect the real-time changes of the scene. Furthermore, by calculating the scene complexity at any moment and combining the complexity of multiple moments to calculate the average complexity, the complexity of the autonomous driving scene can be reflected in real time, providing data support for system decision-making. Finally, the scene is graded according to the average complexity, and the complexity level is intuitively divided, which is convenient for testers to understand and apply. The method has strong scalability, can flexibly adjust the factor screening and weight calculation according to needs, and is suitable for the evaluation of various complex scenes.

[0052] Optionally, the step 101 includes:

[0053] Step 1011, obtaining the road grade, traffic facilities, temporary traffic changes, traffic participants, weather conditions and vehicle status information through vehicle-mounted sensors.

[0054] In the embodiment of the present application, the road grade information is the type of the current road (such as expressway, urban road, rural road, etc.) and the road grade (such as main road, secondary road, branch road, etc.) that can be obtained through GPS and map data. In addition, laser radar and cameras can also identify specific features of the road, such as the number of lanes, lane width, road curvature, etc., to further determine the road grade.

[0055] Traffic facility information refers to the various traffic facilities on the road that can be identified by cameras and lidar, such as signal lights, traffic signs, road barriers (crossbars, grilles, ground piles, etc.), anti-collision guardrails, etc. This information is crucial for autonomous vehicles to understand traffic rules, obey traffic signals, and make correct driving decisions.

[0056] Temporary traffic change information includes temporary traffic changes such as road construction, traffic accidents, traffic control, etc. This information can be perceived in real time by vehicle sensors, and the latest traffic information updates can be obtained through the vehicle communication system or the Internet.

[0057] Traffic participant information is the location and speed information of other road users such as vehicles and pedestrians. This information can be obtained through sensors such as lidar, millimeter-wave radar and cameras, and the accuracy and reliability of the information can be improved through data fusion technology. The behavior patterns and intention prediction of traffic participants are crucial for autonomous vehicles to make safe driving decisions.

[0058] Weather condition information is obtained through on-board weather sensors or combined with GPS and Internet weather data, such as rain, snow, fog, light intensity, etc. This information is of great significance for autonomous vehicles to adjust sensor parameters, optimize algorithm performance, and ensure driving safety.

[0059] The vehicle status information includes the vehicle's position, speed, acceleration, yaw angle, heading angle, etc. This information can be obtained in real time through on-board sensors (such as GPS, inertial measurement unit, etc.) and used in the vehicle's control and decision-making systems.

[0060] Use various on-board sensors installed on autonomous vehicles to obtain key information in autonomous driving scenarios. These sensors include but are not limited to LiDAR, cameras, millimeter-wave radar, ultrasonic sensors, and GPS. Each sensor has its own specific functions and uses, and can collect different types of environmental data.

[0061] Specifically,

[0062] Step 1012, cleaning and preprocessing the road grade, traffic facilities, temporary traffic changes, traffic participants, weather conditions and vehicle status information to obtain the autonomous driving scene data.

[0063] In the embodiment of the present application, data cleaning is to remove noise data, abnormal data, duplicate data, etc. Noise data may be erroneous data caused by sensor failure or environmental factors (such as electromagnetic interference); abnormal data may be extreme data caused by special events (such as traffic accidents); duplicate data may be repeated records caused by excessive sensor sampling frequency or data transmission errors. Through data cleaning, the accuracy and consistency of the data can be ensured.

[0064] Data preprocessing is to format, normalize and extract features from data. Formatting is to convert data into a unified format and unit for subsequent processing and analysis; normalization is to scale the data to a specific range (such as between 0 and 1) to improve the convergence speed and stability of the algorithm; feature extraction is to extract feature information from the original data that is useful for evaluating autonomous driving scenarios, such as the number of lanes, slope, turning curvature, speed of traffic participants, etc.

[0065] Data fusion is the fusion of data from different sensors to improve the accuracy and reliability of information. Data fusion can be achieved through a variety of methods, such as weighted averaging, Kalman filtering, particle filtering, etc. Through data fusion, more comprehensive and accurate autonomous driving scene data can be obtained.

[0066] Clean and preprocess these data to ensure data quality and accuracy. Data cleaning and preprocessing are key links in autonomous driving scenario evaluation, which directly affects the performance of subsequent algorithms and the reliability of results.

[0067] Optionally, the step 102 includes:

[0068] Step 1021, classify the autonomous driving scene data according to the six-level principle, where the six-level principle includes road grade, traffic facilities, temporary traffic changes, traffic participants, weather conditions and vehicle status information.

[0069] In the embodiment of this application, we need to classify the acquired autonomous driving scene data according to the pre-set six-level principle. This six-level principle is based on the ISO 34502 standard and the actual needs of autonomous driving scene evaluation, and aims to comprehensively and systematically consider various factors that affect the complexity of autonomous driving scenes.

[0070] The six-level principle includes: road grade, traffic facilities, temporary traffic changes, traffic participants, weather conditions and vehicle status information. Among them, road grade reflects the basic characteristics and types of roads, such as expressways, urban roads, rural roads, etc., as well as specific road grades (such as main roads, secondary roads, etc.). Traffic facilities include fixed facilities on the road, such as signal lights, traffic signs, road gates (crossbars, grilles, ground piles, etc.), anti-collision guardrails, etc. Temporary traffic changes refer to temporary changes on the road, such as road construction, traffic accidents, traffic control, etc. Traffic participants are other vehicles and pedestrians on the road, and their behavior patterns and intention predictions are crucial to the decision-making of autonomous vehicles. Weather conditions include rain, snow, fog, light intensity, etc., which will affect the sensor performance and algorithm accuracy of autonomous vehicles. The vehicle status refers to the status information of the autonomous vehicle itself, such as position, speed, acceleration, yaw angle, heading angle, etc.

[0071] Each data point is classified into the corresponding level according to its attributes (such as road type, traffic facility type, weather conditions, etc.). For each level, specific elements are further subdivided (such as the number of lanes in the road level, traffic lights in the traffic facilities, etc.).

[0072] Step 1021: Filter out key factors that affect the complexity of the autonomous driving scenario in each level according to the classification results.

[0073] After completing the six-level classification of the data, the next step is to screen out the key factors that affect the complexity of the autonomous driving scene in each level based on the classification results. This step is based on an in-depth understanding and analysis of the complexity of the autonomous driving scene, and aims to identify those factors that have the greatest impact on the performance and safety of autonomous driving vehicles.

[0074] Regarding the screening criteria, the degree of impact on the performance of the autonomous vehicle is to consider whether the factor directly affects the decision-making, control or sensor performance of the autonomous vehicle. The frequency and importance of occurrence are the frequency of the factor in the autonomous driving scene and its impact on the scene complexity. The quantifiability is whether the factor can be quantified for subsequent complexity calculation and evaluation.

[0075] For each level, the importance of each element is evaluated according to the screening criteria. The key elements in each level are screened out, and these elements will be the main objects of subsequent complexity calculation and evaluation. Based on the screening results, a list of elements (as shown in Table 1) can be drawn up, listing the key elements in each level and their types (Boolean or non-Boolean).

[0076] For example, the scene complexity requirements screened out based on existing user big data are as shown in Table 1 below:

[0077] Table 1

[0078]

[0079]

[0080] Among them, the number of lanes: the identification of the number of lanes will affect the identification results of the surrounding vehicles relative to the lane position of the vehicle. When there are multiple lanes, the decision needs to consider which lane to drive on, which affects the task decision. The path planning span of multiple lanes is large, which increases the difficulty compared to the path planning with a small span. Therefore, it will affect L5 target recognition, L4 task decision and L3 path planning.

[0081] Slope: The slope directly affects the vehicle's path tracking ability. The uphill and downhill processes increase the difficulty of path tracking control due to the influence of vehicle gravity. The slope also affects the range of sensor hardware perception, target recognition and road curvature: Road curvature directly affects the path planning function. The greater the curvature of the road, the more curved the path, and the more difficult it is to plan a driving path. Compared with straight roads, it has an impact on path tracking and hardware execution. At the same time, curves will affect the recognition of the relative positions of surrounding target vehicles. Therefore, it will affect L1 hardware execution, L2 path tracking, L3 path planning, L4 task decision-making, and L5 target recognition.

[0082] Barrier type (crossbar): The crossbar, grille, ground stakes and other items in the parking scene will affect the sensor's perception range and affect the parking behavior decision. Therefore, it will affect the L4 task decision and L6 hardware perception.

[0083] Barrier type (barrier): The horizontal bars, grilles, ground stakes and other items in the parking scene will affect the sensor's perception range and affect the parking behavior decision. Therefore, it will affect the L4 task decision and L6 hardware perception.

[0084] Barrier type (ground stake, other): The horizontal bars, grilles, ground stakes and other items in the parking scene will affect the sensor's perception range and affect the parking behavior decision. Therefore, it will affect the L4 task decision and L6 hardware perception.

[0085] Left and right side isolation belt types (cement piers): Concrete piers are road obstructions, which will affect the detection range of the physical obstruction sensor and cause the hardware perception performance to degrade, directly affecting the hardware perception subsystem and recognition system. Therefore, it will affect L5 target recognition and L6 hardware perception.

[0086] Type of left and right side isolation belts (plants): Plants are road occlusions, which will affect the detection range of the physical occlusion sensor and cause the hardware perception performance to degrade, directly affecting the hardware perception subsystem and recognition system. Therefore, it will affect L5 target recognition and L6 hardware perception.

[0087] Types of left and right side isolation belts (shade boards, shading nets): Shade boards and shading nets are road-side shielding, which will affect the detection range of physical shielding sensors and cause hardware perception performance degradation, directly affecting the hardware perception subsystem and recognition system. Therefore, it will affect L5 target recognition and L6 hardware perception.

[0088] Left and right side isolation belt types (pile buckets): Pile buckets are road obstructions that will affect the detection range of the physical obstruction sensor, causing hardware perception performance degradation, and directly affecting the hardware perception subsystem and recognition system. Therefore, it will affect L5 target recognition and L6 hardware perception.

[0089] Type of left and right side isolation belts (anti-collision guardrails): Anti-collision guardrails are road obstructions, which will affect the detection range of physical obstruction sensors and cause hardware perception performance degradation, directly affecting the hardware perception subsystem and recognition system. Therefore, it will affect L5 target recognition and L6 hardware perception.

[0090] Left and right side isolation belt types (other): Other road obstructions will affect the detection range of the physical obstruction sensor, causing hardware perception performance degradation, and directly affecting the hardware perception subsystem and recognition system. Therefore, it will affect L5 target recognition and L6 hardware perception.

[0091] Special road elements (toll booths): Toll booths and other facilities will affect hardware perception, target recognition, and driving path judgment. Therefore, it will affect L4 task decision-making, L5 target recognition, and L6 hardware perception.

[0092] Special road elements (ramps): Ramps are usually blocked by anti-collision guardrails, which will affect the detection range of the sensor and cause the hardware perception performance to degrade, directly affecting the hardware perception subsystem and recognition system. Entering or exiting the ramp will affect the task decision. Therefore, it will affect L4 task decision, L5 target recognition and L6 hardware perception.

[0093] Traffic lights: Traffic lights mainly refer to traffic lights at intersections, which directly affect target recognition and task decision-making. The system must first identify the content of traffic signs. The more types and numbers of traffic signs in the scene, the more difficult it is to identify. The second is task decision-making. The decision-making module must make correct plans based on the content of the traffic signs. Therefore, it will affect L4 task decision-making and L5 target recognition.

[0094] Traffic signs: Traffic signs mainly refer to ground traffic markings and traffic signs on the ground, which directly affect target recognition and task decision-making. The system must first identify the content of the traffic signs. The more types and numbers of traffic signs in the scene, the more difficult it is to identify. The second is task decision-making. The decision-making module must make correct plans based on the content of the traffic signs. Therefore, it will affect L4 task decision-making and L5 target recognition.

[0095] Obstacles (warning signs): Warning signs are road obstructions that will affect the detection range of the physical obstruction sensor, causing the hardware perception performance to degrade, directly affecting the hardware perception subsystem and recognition system. After recognizing the intention of the warning sign, corresponding task decisions and path planning should be made based on the judgment results. Therefore, it will affect L3 path planning, L4 task decision-making, L5 target recognition and L6 hardware perception.

[0096] Obstacles (cones): Cones are road obstructions that affect the detection range of physical obstruction sensors, causing hardware perception performance degradation, and directly affecting the hardware perception subsystem and recognition system. After recognizing the intention of the warning sign, corresponding task decisions and path planning should be made based on the judgment results. Therefore, it will affect L3 path planning, L4 task decision-making, L5 target recognition, and L6 hardware perception.

[0097] Obstacles (other objects): Other objects are road obstructions that will affect the detection range of the physical obstruction sensor, causing the hardware perception performance to degrade, directly affecting the hardware perception subsystem and recognition system. After recognizing the intention of the warning sign, the corresponding task decision and path planning should be made based on the judgment results. Therefore, it will affect L3 path planning, L4 task decision, L5 target recognition and L6 hardware perception.

[0098] Traffic participants: The physical properties of traffic participants are mainly size, material, and motion characteristics. The size attributes directly affect hardware perception and target recognition. The shape of the traffic vehicle will physically block the sensing of autonomous driving. Secondly, the size of the traffic vehicle will affect the difficulty of target recognition. The smaller the vehicle, the greater the difficulty of recognition. Different materials will affect the effect of target recognition. For example, some people wearing strange clothes on the road may cause target recognition failure. The motion characteristics of the vehicle, such as the lane changing, acceleration and deceleration of the traffic vehicle, will affect the vehicle's decision-making. Different traffic participants have different motion uncertainties, which require different decision-making strategies to deal with, increasing the complexity of system control. Therefore, it will affect L1 hardware execution, L2 path tracking, L3 path planning, L4 task decision-making, L5 target recognition and L6 hardware perception.

[0099] Weather (rain): The physical properties of rain are expressed by rainfall. When the rainfall is heavy, raindrops will form rain lines during their falling process, which will block the target and increase the difficulty of target identification. At the same time, raindrops will scatter electromagnetic waves and reduce the detection distance of millimeter-wave radars. The rain lines formed by raindrops will reflect the laser beam of the lidar, reducing the detection distance of the lidar and directly affecting the hardware sensing system. Therefore, rain and snow directly affect the hardware perception system and algorithm recognition system. Rain will also make the road surface slippery and affect the chassis execution. Therefore, it will affect the L1 chassis execution, L5 target recognition and L6 hardware perception.

[0100] Weather (snow): Irregular falling snowflakes are randomly distributed on the image, directly affecting the target recognition subsystem. Snow coverage will also affect the sensor's perception. Snowy days also make the road slippery, affecting chassis performance. Therefore, it will affect L1 chassis performance, L5 target recognition, and L6 hardware perception.

[0101] Weather (fog): Fog directly affects the imaging distance of the camera. At the same time, the fogged target will also affect the accuracy of target recognition, so fog directly affects the hardware perception system and algorithm recognition system. Therefore, it will affect L5 target recognition and L6 hardware perception.

[0102] Weather (dust): Dust will also affect the imaging distance of the camera. At the same time, targets blocked by dust will also affect the accuracy of target recognition. Therefore, dust directly affects the hardware perception system and algorithm recognition system. Therefore, it will affect L5 target recognition and L6 hardware perception.

[0103] PM2.5: PM2.5 directly affects air quality and creates a fog-like effect, which mainly affects hardware perception and target recognition. Therefore, it will affect L5 target recognition and L6 hardware perception.

[0104] Lighting: Light intensity refers to the brightness of light. Light intensity will affect the imaging distance of the camera. As the light intensity decreases, the imaging distance also decreases, until ordinary cameras cannot image scenes at night when there is no light. At the same time, changes in light intensity will also affect the target recognition algorithm. For example, the recognition difficulty and effect of the same target in bright scenes and dim scenes are different. Within a certain light intensity range, the higher the scene brightness, the better the recognition effect. Therefore, light intensity will directly affect the hardware perception system and algorithm recognition system. Therefore, it will affect L5 target recognition and L6 hardware perception.

[0105] Visual interference (backward sunlight): Backward sunlight affects the incident angle. The local illumination changes caused by the incident angle will cause shadows. Shadows often cause errors in image segmentation and recognition, thus affecting the vision-based target recognition algorithm. Therefore, the properties of the illumination incident angle will have a direct impact on the algorithm recognition system. Therefore, it will affect L5 target recognition.

[0106] Visual interference (oncoming lights): Oncoming lights will affect the incident angle. The local illumination changes caused by the incident angle will cause shadows. Shadows often cause errors in image segmentation and recognition, thus affecting the vision-based target recognition algorithm. Therefore, the properties of the illumination incident angle will have a direct impact on the algorithm recognition system. Therefore, it will affect L5 target recognition.

[0107] Visual interference (super bright street lights): Super bright street lights will affect the incident angle. The local illumination changes caused by the incident angle will lead to shadows. Shadows often cause errors in image segmentation and recognition, thus affecting the vision-based target recognition algorithm. Therefore, the properties of the illumination incident angle will have a direct impact on the algorithm recognition system. Therefore, it will affect L5 target recognition.

[0108] Optionally, the step 103 includes:

[0109] Step 1031: construct the judgment matrix according to the influence transmission times of the elements.

[0110] In the embodiments of the present application, the number of influence transmissions refers to the number of direct or indirect influences one element has on another element. In an autonomous driving scenario, some elements may directly affect other elements, while some elements may indirectly affect other elements through intermediate elements. By analyzing the influence relationship between elements, we can determine the number of influence transmissions of each element on other elements, thereby reflecting its relative importance in the complexity of the scenario.

[0111] The judgment matrix is ​​a square matrix whose number of rows and columns is equal to the number of elements. Each element in the matrix represents the relative importance between two elements, usually expressed on a scale of 1-9, where 1 means that the two elements are equally important and 9 means that one element is extremely important than the other. Based on the number of influence transmissions, we can fill in the elements in the judgment matrix. For example, if one element has a large number of influence transmissions on another element, then in the judgment matrix, the element value corresponding to the element will be relatively large.

[0112] Step 1032: Perform a consistency check on the judgment matrix to ensure the consistency of the judgment matrix.

[0113] In the embodiment of the present application, it is checked whether the elements in the judgment matrix are coordinated with each other to avoid logical contradictions or inconsistencies, and to ensure that the judgment matrix can accurately reflect the relative importance relationship between the elements.

[0114] The consistency ratio (CR) is usually used for testing. CR is the ratio of the consistency index (CI) of the judgment matrix to the random consistency index (RI). First, the maximum eigenvalue (λ_max) of the judgment matrix is ​​calculated, and then the CI is calculated according to the formula. Next, the RI value corresponding to the matrix order is found, and finally the CR is calculated. If the CR is less than a certain threshold (such as 0.1), the judgment matrix is ​​considered to have satisfactory consistency; otherwise, the judgment matrix needs to be adjusted or reconstructed.

[0115] Step 1033: Calculate the weight coefficient of each of the elements according to the judgment matrix.

[0116] In the embodiments of the present application, the eigenvector method is generally used to calculate the weight coefficient. First, the eigenvector corresponding to the maximum eigenvalue of the judgment matrix is ​​calculated. Then, the eigenvector is normalized to obtain the weight coefficient of each element.

[0117] The weight coefficient can be used in subsequent complexity calculations as the weight of the factor's contribution to complexity. Through weighted summation and other methods, the overall complexity of the autonomous driving scenario can be calculated, providing quantitative indicators for scenario evaluation and testing.

[0118] For example, the number of scene complexity elements screened with reference to Table 1 reaches 29, and the constructed judgment matrix will have no eigenvalue solution, so a step-by-step calculation method is adopted to divide the elements into 7 parts, and calculate their weights separately, and finally multiply each weight of each part by the proportion of the total scale of the elements in this part, and the weight coefficients of each of the 29 elements can be obtained, as shown in Table 2.

[0119] Table 2

[0120]

[0121]

[0122] Optionally, the step 104 includes:

[0123] Step 1041, for an element of Boolean type, if the element appears in the autonomous driving scenario, the independent complexity of the element is 1, otherwise it is 0.

[0124] In the embodiment of the present application, in the evaluation of the complexity of the autonomous driving scene, the elements can be divided into Boolean type and non-Boolean type. Boolean type elements refer to those elements with only two possible states, namely "appear" or "do not appear". For such elements, the calculation of their independent complexity is relatively simple.

[0125] Boolean elements in autonomous driving scenarios represent the existence or non-existence of a certain condition or event, such as "whether there are pedestrians crossing the road" or "whether there is road construction".

[0126] If the Boolean value element appears in the autonomous driving scenario, that is, its state is "true" or "1", the independent complexity of the element is 1. If the Boolean value element does not appear in the autonomous driving scenario, that is, its state is "false" or "0", the independent complexity of the element is 0.

[0127] This calculation method intuitively reflects the impact of Boolean elements on the complexity of autonomous driving scenarios. When a Boolean element appears, it increases the complexity of the scenario because the autonomous driving vehicle needs to consider the existence of this element and make corresponding decisions.

[0128] Step 1042, for non-Boolean value type elements, the independent complexity of the element is calculated according to the specific value of the element by using a preset complexity calculation formula.

[0129] In the embodiment of the present application, non-Boolean value type elements refer to those elements with continuous or discrete value ranges, such as "vehicle speed", "rainfall in weather conditions", etc. For such elements, the calculation of their independent complexity needs to be based on specific values ​​and preset complexity calculation formulas.

[0130] Non-Boolean elements are represented in autonomous driving scenarios as variables with a certain range of values, which can change continuously (such as vehicle speed) or discretely (such as the number of lanes).

[0131] For non-Boolean elements, the independent complexity needs to be calculated based on their specific values ​​and the preset complexity calculation formula. The complexity calculation formula may be linear, nonlinear, or piecewise, depending on how and to what extent the element affects the complexity of the autonomous driving scenario. For example, for the element of vehicle speed, a piecewise function may be used to calculate its independent complexity: when the vehicle speed is in the low speed range, the complexity is low; when the vehicle speed is in the high speed range, the complexity is high.

[0132] The contribution of non-Boolean elements to the complexity of autonomous driving scenarios can be quantified through the preset complexity calculation formula. This quantification method allows the complexity of different elements to be compared and accumulated, providing a basis for subsequent overall complexity calculation.

[0133] Optionally, the step 105 includes:

[0134] Step 1051, multiplying the independent complexity of each of the elements by the weight coefficient to obtain the weighted complexity of each of the elements.

[0135] In the embodiment of the present application, the independent complexity reflects the direct contribution of a single factor to the complexity of the autonomous driving scenario, and the weight coefficient reflects the importance or influence of the factor in the overall evaluation.

[0136] The independent complexity of each factor is multiplied by its corresponding weight coefficient, and the result is the weighted complexity of the factor. This calculation process is actually weighting the independent complexity of the factor to reflect its relative importance in the overall complexity evaluation.

[0137] Weighted complexity more accurately reflects the actual impact of factors on the complexity of autonomous driving scenarios. By weighting, we can ensure that those factors that have a greater impact on the complexity of the scenario receive greater attention in the overall evaluation.

[0138] Step 1052, adding the weighted complexities of all the elements to obtain the complexity of the autonomous driving scenario at any moment.

[0139] In the implementation of this application, the weighted complexity of all elements is added together to obtain a comprehensive complexity value, which reflects the overall difficulty or challenge of the autonomous driving scene at a certain moment. This comprehensive complexity value can be used to compare the complexity between different scenes, or to evaluate the complexity changes of the same scene at different times.

[0140] The complexity of autonomous driving scenarios can be used to guide the development and testing of autonomous driving systems. By understanding the complexity of scenarios, we can better design algorithms, optimize system performance, and ensure that autonomous driving systems can operate safely and effectively in various complex scenarios.

[0141] During the addition process, you need to ensure that the weighted complexity of all elements is calculated correctly and there are no omissions or duplications. If the scene contains multiple elements, and these elements have different degrees of influence on complexity, then the weighted complexity addition process will accurately reflect this difference.

[0142] Optionally, the step 106 includes:

[0143] Step 1061: Add the complexity of the autonomous driving scenario at multiple moments to obtain the total complexity of the autonomous driving scenario.

[0144] In the implementation of this application, the complexity of the autonomous driving scene at different times may vary due to changes in traffic conditions, weather conditions, road environment, etc. By calculating the complexity at multiple times, we can have a more comprehensive understanding of the complexity of the scene over a period of time.

[0145] The complexity value of the autonomous driving scenario at each moment is added up to get the total complexity of the scenario over a period of time. This total complexity value reflects the cumulative effect of the complexity of the scenario at all moments during this period of time.

[0146] Total complexity can be used to compare the complexity of different scenes in the same time period. It can also be used to evaluate the changes in complexity of the same scene in different time periods, so as to understand the dynamic characteristics of scene complexity.

[0147] Step 1062: Divide the total complexity by the number of the multiple moments to obtain the average complexity of the autonomous driving scenario.

[0148] In the embodiment of the present application, the average complexity reflects the average complexity of the autonomous driving scene over a period of time, which eliminates the influence of time factors on the total complexity. By calculating the average complexity, we can more accurately understand the complexity of the scene under typical or average conditions.

[0149] The total complexity is divided by the number of moments, and the result is the average complexity of the autonomous driving scene. This calculation process is actually averaging the total complexity to get the average complexity of each moment.

[0150] Average complexity can be used to design and optimize autonomous driving systems to ensure that the system can operate safely and effectively under average complexity conditions. It can also be used to compare the average complexity between different scenarios, so as to select scenarios that are more suitable for testing and verification of autonomous driving systems.

[0151] Specifically, refer to Figure 3 As mentioned above, the complete calculation process of the total scene complexity is as follows:

[0152] The calculation model used to calculate the scene complexity is shown in formula (1) and formula (2). First, the independent complexity corresponding to each scene complexity factor and their respective weights are calculated, and finally the complexity of the scene at a certain moment is obtained by adding them up in proportion. Then, the average complexity of each moment in the scene is taken as the scene complexity. A total of 29 scene complexity factors were screened out through analysis.

[0153]

[0154] Where:

[0155] CTotal Represents the complexity of the scene at a certain moment;

[0156] ω i Represents the weight of each scene complexity factor;

[0157] C i Represents the complexity of each element;

[0158] i represents the number of scene complexities involved in the calculation;

[0159] SC represents scene complexity;

[0160] n represents the total number of moments of complexity contained in this scene.

[0161] Quantification of scene element complexity:

[0162] 1. Boolean type elements

[0163] For elements of Boolean type, if they appear in the scene, their own complexity C i That is 1, otherwise C i is 0.

[0164] 2. Non-Boolean type elements

[0165] For non-Boolean elements, such as lighting, number of lanes, slope, etc., it is necessary to bring the specific value of the element into its own complexity formula for calculation to obtain its own complexity:

[0166] PM2.5:

[0167]

[0168] Where:

[0169] C PM2.5 Represents the complexity of the element “PM 2.5” itself;

[0170] n represents the empirical coefficient;

[0171] PM2.5 represents the concentration of PM2.5;

[0172] (2) Lighting:

[0173]

[0174] Where:

[0175] C illumination Represents the complexity of the element "lighting" itself;

[0176] n represents the empirical coefficient;

[0177] web_illumination represents the light intensity;

[0178] (3) Slope:

[0179] Using the altitude at each moment and the distance traveled by the vehicle, a trigonometric function relationship is formed to calculate the lane slope gra as shown in the following formula:

[0180]

[0181] Where:

[0182] gra represents the road slope;

[0183] ΔS is the vehicle's displacement every 0.1 seconds;

[0184] ΔAltitude represents the altitude difference of the vehicle every 0.1 seconds;

[0185]

[0186] Where:

[0187] C slope The complexity corresponding to the representative element "road slope";

[0188] n represents the empirical coefficient;

[0189] gra represents the road slope;

[0190] (4) Curvature:

[0191]

[0192] Where:

[0193] C curvature Represents the complexity corresponding to the element "road curvature";

[0194] n represents the empirical coefficient;

[0195] Curvature represents the curvature of the road and is expressed by the curvature of the vehicle.

[0196] (5) Traffic participants:

[0197] If the x-speed of the vehicle is less than the x-speed of the target vehicle, ttc_x is recorded as 100s. If the x-speed of the vehicle is greater than the x-speed of the target vehicle, ttc_x is calculated as:

[0198]

[0199] Where:

[0200] ttc_x represents the longitudinal collision time between the vehicle and the preceding vehicle;

[0201] obj_posx represents the relative position of the target vehicle and the host vehicle in the x direction;

[0202] vehicle_vehiclespeed represents the vehicle speed;

[0203] vehicle_gnss_orientation represents the vehicle's heading angle;

[0204] obj_vxabs represents the absolute value of the target vehicle's x-speed.

[0205] If the distance between the target vehicle and the host vehicle in the y direction is increasing, ttc_ is recorded as 100s; if the host vehicle and the target vehicle are traveling in opposite directions, ttc_y is calculated as:

[0206]

[0207] Where:

[0208] ttc_y represents the lateral collision time between the vehicle and the preceding vehicle;

[0209] obj_posy represents the relative position of the target vehicle and the host vehicle in the y direction;

[0210] vehicle_vehiclespeed represents the vehicle speed;

[0211] vehicle_gnss_orientation represents the vehicle's heading angle;

[0212] obj_vyabs represents the absolute value of the target vehicle's y-speed.

[0213] If the host vehicle and the target vehicle are traveling in the same direction, calculate ttc_y:

[0214]

[0215] Where:

[0216] ttc_y represents the lateral collision time between the vehicle and the preceding vehicle;

[0217] obj_posy represents the relative position of the target vehicle and the host vehicle in the y direction;

[0218] vehicle_vehiclespeed represents the vehicle speed;

[0219] vehicle_gnss_orientation represents the vehicle's heading angle;

[0220] obj_vyabs represents the absolute value of the target vehicle's y-speed.

[0221] Finally, the calculation logic of traffic participant information complexity is as follows:

[0222]

[0223] Where:

[0224] ttc_xi represents the longitudinal collision time between the host vehicle and the i-th target vehicle;

[0225] ttc_yi represents the lateral collision time between the vehicle and the i-th target vehicle;

[0226] Ntp represents the number of traffic participants other than the vehicle itself;

[0227] CTP represents the complexity of traffic participants other than the vehicle itself.

[0228] The vehicle information considers the impact of the vehicle speed, lateral and longitudinal acceleration, and yaw rate on the danger level of the scene. The calculation results of the vehicle information complexity are as follows:

[0229]

[0230] Where:

[0231] Δax represents the lateral acceleration of the vehicle;

[0232] Δay represents the lateral acceleration of the vehicle;

[0233] ω θ Represents the vehicle's heading angular velocity;

[0234] CSV represents the complexity of the vehicle's lateral and longitudinal accelerations and heading angles.

[0235]

[0236] Where:

[0237] v represents the vehicle speed;

[0238] V max Represents the maximum speed of the vehicle in each session;

[0239] V Vehicle_speed Represents the complexity corresponding to the vehicle speed.

[0240] Weight calculation method: Hierarchy analysis method. Hierarchy analysis method is an effective means to solve complex evaluation problems. After analyzing the influencing factors and internal relationships, a hierarchical model is constructed, and then the influence weight coefficient ω of the influencing elements is obtained by using the hierarchical structure process. i .

[0241] Calculation of scene element complexity weights: Solve for the maximum eigenvalue λ of the judgment matrix max And the eigenvector E=(e 1 ,e 2 ,...,e n ), and normalize the feature vector. The resulting vector is the weight coefficient ωi of each scene element. Considering that the autonomous driving system is a multi-level series subsystem, the impact of scene elements on the autonomous driving system will be transmitted step by step. The transmission mechanism is as follows: Figure 2 shown.

[0242] Divide the system into levels L1 to L6, and the number of transmissions Q affected by elements at different levels is:

[0243]

[0244] Where n is the number of attributes of scene elements, L j It is the system level directly affected by the scene element attributes.

[0245] According to the number of transmissions of different elements, the corresponding counting scale αi is then calculated for subsequent weight calculation. There are 29 scene elements considered this time, and their weight coefficients are determined according to the number of transmissions.

[0246] Create a judgment matrix: Calculate the number of scene element influence transfers and their corresponding scale values ​​α according to the hierarchical model i , and based on this, establish the corresponding judgment matrix.

[0247]

[0248] Scenario complexity classification: Although detailed quantitative indicators are helpful for analyzing the complexity of the scenario itself, in actual applications, it is not conducive to the tester's intuitive understanding of the complexity of the scenario. Therefore, based on the calculated quantitative indicators, the complexity of the scenario is divided into 5 levels, each of which can represent a different level of complexity, making the design and test application purpose of the scenario clearer.

[0249] The scene complexity calculation model studied in this paper selects appropriate normalization methods and weight indicators for each evaluation dimension, and the final scene complexity quantification value is a continuous quantity in the interval (0,1). To divide it into 6 levels, we can use the equal interval division method.

[0250] First, we need to determine the boundary value of each level. Since it is from 0 to 1, there are 6 levels in total, we can divide this range into 6 equal parts, and the interval of each equal part is 0.166. The following are the 6 levels divided according to this method as shown in Table 3:

[0251] Table 3

[0252] Grading Complexity Range Low complexity (0,0.166) Low to medium complexity [0.166,0.332) Medium complexity [0.332,0.498) Medium to high complexity [0.498,0.664) High complexity [0.664,0.830) Extremely high complexity [0.830，1)

[0253] Based on the same inventive concept, the embodiment of the present application also provides an autonomous driving benchmark scenario library complexity evaluation device for implementing the autonomous driving benchmark scenario library complexity evaluation method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more autonomous driving benchmark scenario library complexity evaluation device embodiments provided below can refer to the limitations of the autonomous driving benchmark scenario library complexity evaluation method above, and will not be repeated here.

[0254] In an exemplary embodiment, Figure 4 As shown, a device 20 for evaluating the complexity of an autonomous driving benchmark scenario library is provided, comprising:

[0255] An acquisition module 201 is used to acquire the autonomous driving scene data to be evaluated;

[0256] An evaluation module 202, configured to screen out factors that affect the complexity of the autonomous driving scenario according to preset standards;

[0257] Based on the hierarchical analysis method, a judgment matrix of the elements is constructed, and a weight coefficient of each element is calculated according to the judgment matrix;

[0258] According to the type of the element, the independent complexity of each element is calculated respectively, and the type of the element includes a Boolean value type and a non-Boolean value type;

[0259] Calculating the complexity of the autonomous driving scenario at any moment according to the independent complexity of the elements and the weight coefficients;

[0260] Calculating an average complexity of the autonomous driving scenario according to the complexity of the autonomous driving scenario at multiple moments;

[0261] The autonomous driving scenarios are graded according to the average complexity to obtain complexity levels of the autonomous driving scenarios.

[0262] Optionally, the acquisition module 201 is further used to:

[0263] Acquire the road grade, traffic facilities, temporary traffic changes, traffic participants, weather conditions and vehicle status information through vehicle-mounted sensors;

[0264] The road grade, traffic facilities, temporary traffic changes, traffic participants, weather conditions and vehicle status information are cleaned and preprocessed to obtain the autonomous driving scene data.

[0265] Optionally, the evaluation module 202 is further used to:

[0266] Classifying the autonomous driving scenario data according to the six-level principle, wherein the six-level principle includes road grade, traffic facilities, temporary traffic changes, traffic participants, weather conditions, and vehicle status information;

[0267] According to the classification results, key factors affecting the complexity of the autonomous driving scenario at each level are screened out.

[0268] Optionally, the evaluation module 202 is further used to:

[0269] Constructing the judgment matrix according to the number of times the influence of the elements is transmitted;

[0270] Performing a consistency check on the judgment matrix to ensure the consistency of the judgment matrix;

[0271] According to the judgment matrix, the weight coefficient of each of the elements is calculated.

[0272] Optionally, the evaluation module 202 is further used to:

[0273] For an element of a Boolean type, if the element appears in the autonomous driving scenario, the independent complexity of the element is 1, otherwise it is 0;

[0274] For elements of non-Boolean value type, the independent complexity of the element is calculated according to the specific value of the element using a preset complexity calculation formula.

[0275] Optionally, the evaluation module 202 is further used to:

[0276] Multiplying the independent complexity of each of the elements by the weight coefficient to obtain the weighted complexity of each of the elements;

[0277] The weighted complexity of all the elements is added together to obtain the complexity of the autonomous driving scenario at any moment.

[0278] Optionally, the evaluation module 202 is further used to:

[0279] Adding the complexity of the autonomous driving scenario at multiple moments to obtain the total complexity of the autonomous driving scenario;

[0280] The total complexity is divided by the number of the multiple moments to obtain the average complexity of the autonomous driving scenario.

[0281] The embodiment of the present application obtains multi-dimensional autonomous driving scene data and selects complexity factors in combination with the six-level principle in the preset standard regulations to ensure the comprehensiveness and standardization of the evaluation. The judgment matrix is ​​constructed and the weight coefficient is calculated using the hierarchical analysis method to finely quantify the impact of each factor on the scene complexity, avoiding the subjectivity of weight allocation in traditional methods. By calculating the independent complexity of Boolean and non-Boolean elements respectively, the actual impact of different elements is dynamically reflected, especially the complexity calculation of non-Boolean elements, which can more accurately reflect the real-time changes of the scene. Furthermore, by calculating the scene complexity at any moment and combining the complexity of multiple moments to calculate the average complexity, the complexity of the autonomous driving scene can be reflected in real time, providing data support for system decision-making. Finally, the scene is graded according to the average complexity, and the complexity level is intuitively divided, which is convenient for testers to understand and apply. The method has strong scalability, can flexibly adjust the factor screening and weight calculation according to needs, and is suitable for the evaluation of various complex scenes.

[0282] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 5 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. Among them, the processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store complexity evaluation data of an autonomous driving benchmark scenario library. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for evaluating the complexity of an autonomous driving benchmark scenario library is implemented.

[0283] Those skilled in the art will understand that Figure 5 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 those shown in the figure, or combine certain components, or have a different arrangement of components.

[0284] In an exemplary embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0285] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0286] In an exemplary embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0287] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0288] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and 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 embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present 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 may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0289] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.

[0290] The technical features of the above embodiments may 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.

[0291] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A complexity evaluation method for an autonomous driving benchmark scenario library, characterized in that: The complexity evaluation method of the autonomous driving benchmark scenario library includes: Obtain the autonomous driving scenario data to be evaluated; Screening out factors that affect the complexity of the autonomous driving scenario according to preset standard regulations; Based on the hierarchical analysis method, a judgment matrix of the elements is constructed, and a weight coefficient of each element is calculated according to the judgment matrix; According to the type of the element, the independent complexity of each element is calculated respectively, and the type of the element includes a Boolean value type and a non-Boolean value type; Calculating the complexity of the autonomous driving scenario at any moment according to the independent complexity of the elements and the weight coefficients; Calculating an average complexity of the autonomous driving scenario according to the complexity of the autonomous driving scenario at multiple moments; The autonomous driving scenarios are graded according to the average complexity to obtain complexity levels of the autonomous driving scenarios.

2. The complexity evaluation method of the autonomous driving benchmark scenario library according to claim 1 is characterized in that: The step of obtaining the autonomous driving scene data to be evaluated includes: Acquire the road grade, traffic facilities, temporary traffic changes, traffic participants, weather conditions and vehicle status information through vehicle-mounted sensors; The road grade, traffic facilities, temporary traffic changes, traffic participants, weather conditions and vehicle status information are cleaned and preprocessed to obtain the autonomous driving scene data.

3. The complexity evaluation method of the autonomous driving benchmark scenario library according to claim 1, characterized in that: The step of screening out the factors affecting the complexity of the autonomous driving scenario according to the preset standard regulations includes: Classifying the autonomous driving scenario data according to the six-level principle, wherein the six-level principle includes road grade, traffic facilities, temporary traffic changes, traffic participants, weather conditions, and vehicle status information; According to the classification results, key factors affecting the complexity of the autonomous driving scenario at each level are screened out.

4. The complexity evaluation method of the autonomous driving benchmark scenario library according to claim 1, characterized in that: The step of constructing a judgment matrix of the elements based on the hierarchical analysis method and calculating the weight coefficient of each element according to the judgment matrix includes: Constructing the judgment matrix according to the number of times the influence of the elements is transmitted; Performing a consistency check on the judgment matrix to ensure the consistency of the judgment matrix; According to the judgment matrix, the weight coefficient of each of the elements is calculated.

5. The complexity evaluation method of the autonomous driving benchmark scenario library according to claim 1, characterized in that: The step of calculating the independent complexity of each of the elements according to the type of the elements comprises: For an element of a Boolean type, if the element appears in the autonomous driving scenario, the independent complexity of the element is 1, otherwise it is 0; For elements of non-Boolean value type, the independent complexity of the element is calculated according to the specific value of the element using a preset complexity calculation formula.

6. The complexity evaluation method of the autonomous driving benchmark scenario library according to claim 1, characterized in that: The step of calculating the complexity of the autonomous driving scenario at any time according to the independent complexity of the elements and the weight coefficients comprises: Multiplying the independent complexity of each of the elements by the weight coefficient to obtain the weighted complexity of each of the elements; The weighted complexity of all the elements is added together to obtain the complexity of the autonomous driving scenario at any moment.

7. The complexity evaluation method of the autonomous driving benchmark scenario library according to claim 1, characterized in that: The step of calculating the average complexity of the autonomous driving scenario according to the complexity of the autonomous driving scenario at multiple moments includes: Adding the complexity of the autonomous driving scenario at multiple moments to obtain the total complexity of the autonomous driving scenario; The total complexity is divided by the number of the multiple moments to obtain the average complexity of the autonomous driving scenario.

8. An autonomous driving benchmark scenario library complexity evaluation device, characterized in that: The autonomous driving benchmark scenario library complexity evaluation device comprises: An acquisition module, used to acquire the autonomous driving scene data to be evaluated; An evaluation module, used to screen out factors that affect the complexity of the autonomous driving scenario according to preset standard regulations; Based on the hierarchical analysis method, a judgment matrix of the elements is constructed, and a weight coefficient of each element is calculated according to the judgment matrix; According to the type of the element, the independent complexity of each element is calculated respectively, and the type of the element includes a Boolean value type and a non-Boolean value type; Calculating the complexity of the autonomous driving scenario at any moment according to the independent complexity of the elements and the weight coefficients; Calculating an average complexity of the autonomous driving scenario according to the complexity of the autonomous driving scenario at multiple moments; The autonomous driving scenarios are graded according to the average complexity to obtain complexity levels of the autonomous driving scenarios.

9. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the complexity evaluation method for the autonomous driving benchmark scenario library according to any one of claims 1 to 7.

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 complexity evaluation method of the autonomous driving benchmark scenario library described in any one of claims 1-7 are implemented.

Images