Key Quantitative Methods for Integrating Subjective and Objective Tests in Complex Testing Scenarios for Intelligent Vehicles

Abstract

This invention relates to a subjective and objective fusion criticality quantification method for complex testing scenarios of intelligent vehicles. The method includes processing a complex testing scenario dataset, employing a subjective and objective fusion criticality quantification method, and using a soft clustering algorithm to quantify the criticality of the testing scenarios. First, the method integrates subjective driver evaluation and objective risk field quantification to assess the criticality of complex testing scenarios. Second, the soft clustering method addresses the imbalance in sample size, improving efficiency while reducing labor costs. This invention can determine the testing value and necessity of complex testing scenarios and identify scenarios with high testing value. This invention provides a research foundation for the extraction and generation of critical testing scenarios.

Description

Technical Field

[0001] This invention relates to a method for extracting and generating key test scenarios for intelligent vehicles, and particularly to a method for quantifying key aspects of complex test scenarios for intelligent vehicles by integrating subjective and objective factors. Background Technology

[0002] Intelligent vehicles represent a crucial direction for the future development of the automotive industry. They achieve intelligent driving tasks through perception, decision-making, planning, and control systems. However, real-world driving environments, including complex weather and traffic conditions, pose significant challenges to the safe operation of intelligent vehicles. Therefore, before the actual deployment of intelligent vehicles, thorough testing of their performance in complex environments is essential to ensure the normal operation of their functions. Scenario-based testing has become a vital method, improving testing efficiency and shortening the development cycle. However, the number of complex test scenarios collected in practice is vast. Using only these scenarios would waste significant computing resources and incur high testing costs. Therefore, there is an urgent need to establish a key quantification method for complex test scenarios of intelligent vehicles. Existing research methods mostly employ objective key quantification methods, such as collision time and headway. However, even after testing intelligent vehicles using key test scenarios selected by these methods, omissions of key test scenarios may still occur, reducing driver acceptance of intelligent vehicle functions. Summary of the Invention

[0003] To address the aforementioned technical problems, this invention provides a key quantitative method for the fusion of subjective and objective methods in complex testing scenarios for intelligent vehicles, comprising the following steps:

[0004] (1) Processing complex test scenario datasets:

[0005] The data at time t of the nth complex test scenario is split into three scenario elements, including meteorological elements. Transportation elements and road elements

[0006] Among them, meteorological elements Including rainfall intensity R a (Unit: mm / h) and fog visibility Vis (unit: m); traffic elements This includes the vehicle's lateral and longitudinal positions (in meters), speed (in meters per second), and acceleration (in meters per second). 2 ); road elements This includes the road surface adhesion coefficient μ, lane line position (unit: m), and lane line linearity (including solid lines and dashed lines);

[0007] (2) Key quantitative method integrating subjective and objective factors

[0008] (2.1) Key Quantification Methods for Subjective and Complex Test Scenarios

[0009] First, virtual simulation software is used to reproduce the three scene elements mentioned above, that is, to construct a complex scene in the virtual simulation scene that is the same as the real environment.

[0010] The driving perspective of the main vehicle is then displayed on the driving simulator, which also includes rearview and side mirrors, allowing the person performing the subjective quantification to fully perceive the behavior of vehicles around the main vehicle.

[0011] Several drivers were recruited, and after observing videos on a driving simulator, a complex scenario evaluation, i.e., criticality quantification, was required for the test scenario. The subjective evaluation matrix focused on two aspects of the test scenario: the degree to which the driver believed there was a collision risk in the scenario and whether the driver was willing to control the vehicle to improve traffic efficiency. The subjective evaluation matrix was used to determine the subjective evaluation results corresponding to the complex test scenario.

[0012] This invention assigns greater weight to the risk dimension in the evaluation matrix. To ensure the rationality of subjective quantification, several different drivers are selected to quantify each test scenario, and the subjective quantification result Lab for the nth test scenario is obtained by weighting the results based on the number of tests. n The specific weighting process is as follows:

[0013]

[0014] In the formula, j represents the scene label value; m j Select the number of people corresponding to the j label for each driver.

[0015] Using only subjective methods for quantifying the criticality of complex test scenarios can lead to a decrease in critical test scenario coverage. Therefore, it is necessary to use objective methods for quantifying the criticality of complex test scenarios to obtain a more comprehensive picture of critical test scenarios.

[0016] (2.2) Critical Quantification Methods for Objective and Complex Test Scenarios

[0017] The key quantification method for objective and complex test scenarios is constructed based on the risk field method. The risk field method is used to construct risk fields for traffic elements and road elements respectively, and to construct field strength attenuation coefficients using meteorological elements. Then, the risk fields generated by the main vehicle and the environment are constructed using the above field strengths respectively, and the interaction between them is used to construct interactive field strength features. The interactive field strength feature (IFF) is defined as the quantification result of the objective method, which integrates the three heterogeneous elements of traffic, road and meteorological elements of the scenario into a single quantification result, thereby objectively describing the driving risk of the main vehicle in the current driving environment.

[0018] The calculation process for the interactive field features is as follows:

[0019] First, the road potential field is generated by road elements, traffic elements, and meteorological elements respectively. Traffic situation and adverse weather field intensity degradation factors Based on this, the environmental field strength was constructed. Then, according to The same method is used to calculate the risk field generated by the main vehicle. Finally, discretization is performed based on the Delaunay triangular discretization method. and Given two continuous non-uniform potential fields, calculate the interaction value IFF between the two discrete potential fields. n,t ; and The stronger the repulsion between them, the stronger the corresponding IFF. n,t The larger the value, the greater the safety risk to the vehicle posed by the driving environment at time t in the nth scenario; the maximum IFF value is taken from all times in the scenario. n,t As a key quantitative result in objective and complex testing scenarios, IFF n To ensure the consistency of the dimensions of key quantitative results, the IFF will be used. n Normalization is performed.

[0020] Furthermore, in step (2.2), the traffic potential field... The calculation method is as follows:

[0021] Traffic potential field generated by a single vehicle m Defined as a representation of the collision risk between itself and its surroundings based on motion parameters such as position, velocity, and acceleration, its mapping relationship is expressed as follows:

[0022]

[0023] The expression is:

[0024]

[0025] Among them, M m The virtual mass is determined by the type and actual mass of vehicle m; γ R This describes the change in the road surface adhesion coefficient μ caused by rainwater. r c represents the coefficient of friction reduction on the road surface after rainfall. r It is a constant that determines the risk growth rate as it approaches m, c r Set it to 0.7; It is the spatiotemporal distance between any point in the spatiotemporal domain and m, and its construction process takes into account vehicle speed.

[0026] The road potential field The calculation method is as follows:

[0027] Road potential field Two important road elements are mainly considered: one is the potential field generated by static obstacles. Such as obstacles; another type is the potential field generated by the lane. The resulting risks are uniformly distributed, in calculation When the speed is set to 0, the result is obtained corresponding to a static obstacle. For a lane, the potential field generated by the lane The expression is:

[0028]

[0029] Where, N L Indicates the number of lanes; I i The meaning of the lane line linearity ε for the i-th lane is that the value corresponding to the solid line is greater than the value corresponding to the dashed line; σ is a fixed value, the magnitude of which determines when the vehicle approaches the lane line. growth rate; The expression is:

[0030]

[0031] The intelligent vehicle perception system for weather conditions mainly considers rain and fog, modeling the impact of rain and fog on the vehicle's perception module; and assessing the adverse weather field intensity degradation factors generated by rain and fog on the vehicle's perception module. Affected by meteorological factors The coupling effect of the longitudinal distance Dis between the target and the vehicle is adopted. This indicates that the choice of W depends on the calibration of different sensors; This represents the risk amplification factor for the scenario compared to the case without adverse weather conditions. The modeling process is to determine Dis and The variation pattern of the intensity degradation factor under different combinations.

[0032] Furthermore, The construction method was applied to the YOLO-v5 target recognition function in the camera perception module of an intelligent vehicle under rainy and foggy weather; in order to obtain Dis and The combined parameters are used to construct W by generating images using PreScan and Simulink; firstly, different parameters are constructed in PreScan. The system first identifies different Dis values ​​in various scenarios and records the current Dis value and corresponding image using Simulink. Then, it inputs the acquired image into YOLO-v5 to identify targets in the image and records the Dis values. The IOU recognition results under different combinations are defined as IOU. W Again, no record. The IOU recognition results under different Dis are defined as IOU. O Finally, Dis, and perceived attenuation coefficient Θ W Represented by the x-axis, y-axis, and z-axis respectively;

[0033] IOU W and IOU O The absolute difference between them is defined as Θ W Θ was obtained respectively in the same Dis scene under the influence of weather conditions and without the influence of weather conditions. W The expression is:

[0034] Θ W =IOU W -IOU O (8)

[0035] Records included rainfall intensity R a Dis and fog visibility (Vis) Different combinations of IOU W As a result, Θ is constructed. W ; The expression is:

[0036]

[0037] Where, β T Θ is the object category coefficient, used to determine the degree of risk posed by different target objects to the vehicle. It is important to emphasize that Θ... W It is a function of W, and can also be calibrated by other intelligent vehicle perception systems.

[0038] Based on the three potential field calculation methods described above, the potential field at time t in scenario n is obtained. expression:

[0039]

[0040] for The establishment of its value is determined by and The maximum strength between and multiplied by Sure.

[0041] Furthermore, in step (2.2), the interaction value IFF between the two discrete potential fields... n,t The calculation method is as follows:

[0042] In obtaining and Next, the continuous potential field is first discretized using the Delaunay triangle discretization method; then, the IFF is calculated using the feature points of each triangular surface. n,t ;

[0043] The Delaunay triangle has advantages such as uniqueness, optimality, and closest proximity. It is used to process... and The corresponding triangular surface is obtained. and By using Risk characteristics and Risk characteristics Quantification separately and IFF between n,t ; and The calculation method is the same, with Let's take the i-th example as an example: The coordinates of the three vertices are respectively and Correspondingly The expression is:

[0044]

[0045] According to the law of universal gravitation, IFF n,t Depend on and Quantization, its expression is:

[0046]

[0047] in, It refers to the i-th scene in the n-th scene. The coordinates at time t; It refers to the j-th scene in the n-th scene. The coordinates at time t.

[0048] (2.3) The final quantitative result C of the scenario is obtained by averaging the key quantitative results of subjective and objective test scenarios. n Its expression is:

[0049]

[0050] (3) Use soft clustering algorithm to quantify the criticality of test scenarios.

[0051] To improve the efficiency of criticality quantification, a soft clustering algorithm is used to perform criticality quantification on unquantized test scenarios. The soft clustering algorithm takes into account the imbalance of sample numbers. Based on the quantified criticality results, the soft clustering algorithm is used to automatically perform criticality quantification on other scenarios, and output the final criticality quantification results of complex test scenarios.

[0052] Soft clustering algorithms address the strict membership relationships of traditional hard clustering algorithms, where membership values ​​are either 0 or 1. Using fuzzy set theory, they expand the original membership values ​​to any value between 0 and 1. A sample can belong to different clusters with different membership values, thus greatly improving the clustering algorithm's ability to handle real-world datasets. The Generalequalization Fuzzy C-means Clustering Algorithm (GEFCM) improves upon the optimization criteria of the traditional fuzzy C-means clustering algorithm by considering sample size in the objective function design. This effectively alleviates the inaccuracy of clustering results caused by the small proportion of key scenarios in practical applications. The objective function of GEFCM is:

[0053]

[0054] And the following constraints are met:

[0055]

[0056] In the formula, m represents the ambiguity control coefficient; c i x represents the cluster center of the i-th cluster (i = 1, 2, ..., c), where c represents the total number of clusters; j For test scenario data, j = 1, 2, ..., n represents a total of n test scenario data to be clustered; U = (u ij ) c*n Let u be the membership matrix. ij The membership degree is represented by x. j The degree to which a cluster belongs to the i-th cluster; Dis(x)j ,c i ) represents scenario x j With scene c i The similarity between them needs to be considered separately for traffic N in the scenario. T Road N R and meteorology N W The parameter values ​​are specifically defined as follows:

[0057]

[0058] In the formula, This represents the average DTW distance between the vehicle trajectories at all corresponding locations in the two scenarios; a new objective function is constructed based on the GEFCM objective function and constraints using the Lagrange multiplier method:

[0059]

[0060] For u in equation (16) ij and c i Taking the partial derivative and setting the result to 0, we obtain:

[0061]

[0062] By summing i from both sides, we get:

[0063]

[0064] Thus, we obtain the Lagrange multiplier λ. j The expression:

[0065]

[0066] λ j Substituting into the original expression, we get u ij Iterative formula:

[0067]

[0068] Soft clustering algorithms can be used to obtain quantitative results for test scenarios that have not undergone subjective and objective key quantification. ij and c i The expression for u is a series of coupled equations, and an analytical solution cannot be obtained. Therefore, u is obtained through iteration. ij and c i The estimated value.

[0069] The beneficial effects of this invention are:

[0070] This invention proposes a subjective and objective quantification method for criticality in complex testing scenarios of intelligent vehicles. First, this method integrates subjective driver evaluation and objective risk field quantification to determine the criticality of complex testing scenarios. Second, it utilizes a soft clustering method to account for imbalanced sample sizes, improving efficiency while reducing labor costs. The subjective quantification of criticality in complex testing scenarios offers two advantages: firstly, it allows drivers to comprehensively analyze the challenges posed by complex environments to driving tasks, thereby identifying challenging testing scenarios; secondly, using these driver-selected scenarios to test intelligent vehicles increases driver acceptance if the vehicles can fully handle the selected scenarios. However, relying solely on subjective quantification leads to a decrease in critical testing scenario coverage. Therefore, this invention employs an objective quantification method to obtain more comprehensive critical testing scenarios. This objective method is constructed based on a risk field approach, which integrates three heterogeneous elements—traffic, road, and weather—into a single quantification result. This invention can determine the testing value and necessity of complex testing scenarios and identify scenarios with high testing value. This invention can establish a research foundation for the extraction and generation of key test scenarios. Attached Figure Description

[0071] Figure 1 This is a schematic diagram of the key quantitative method for the integration of subjective and objective methods for complex testing scenarios of intelligent vehicles, as described in this invention.

[0072] Figure 2 This is a schematic diagram illustrating the use of virtual simulation software to reproduce complex test scenarios according to the present invention.

[0073] Figure 3 This is a schematic diagram of the subjective key quantification matrix of the present invention.

[0074] Figure 4 This is a schematic diagram of the interactive field feature construction process of the present invention.

[0075] Figure 5 To improve the performance of the YOLO-v5 target recognition function in different d W and A schematic diagram of the acquisition process under the combination.

[0076] Figure 6 Different Dis and A schematic diagram showing the sensing attenuation coefficient results of YOLO-v5 under combined conditions.

[0077] Figure 7 For the present invention Schematic diagram of the construction method process.

[0078] Figure 8 This is a schematic diagram illustrating the interactive field eigenvalue calculation process between two discrete potential fields according to the present invention.

[0079] Figure 9 This is a schematic diagram of the key quantification results obtained by the soft clustering algorithm in this invention. Detailed Implementation

[0080] This embodiment provides a key quantitative method that integrates subjective and objective data for complex testing scenarios of intelligent vehicles, such as... Figure 1 As shown, it includes the following steps:

[0081] (1) Processing complex test scenario datasets:

[0082] The data at time t of the nth complex test scenario is split into three scenario elements, including meteorological elements. Transportation elements and road elements

[0083] Among them, meteorological elements Including rainfall intensity R a (Unit: mm / h) and fog visibility Vis (unit: m); traffic elements This includes the vehicle's lateral and longitudinal positions (in meters), speed (in meters per second), and acceleration (in meters per second). 2 ); road elements This includes the road surface adhesion coefficient μ, lane line position (unit: m), and lane line linearity (including solid lines and dashed lines);

[0084] (2) Key quantitative method integrating subjective and objective factors

[0085] (2.1) Key Quantification Methods for Subjective and Complex Test Scenarios

[0086] First, virtual simulation software is used to reproduce the three scene elements mentioned above, that is, to construct a complex scene in the virtual simulation scene that is the same as the real environment.

[0087] Specific steps are as follows: Figure 2 As shown, weather and road elements need to be preset in PreScan. For traffic elements, the FromSpreadsheet module in Simulink needs to be used to read vehicle motion data from each frame.

[0088] The driving perspective of the main vehicle is then displayed on the driving simulator, which also includes rearview and side mirrors, allowing the person performing the subjective quantification to fully perceive the behavior of vehicles around the main vehicle.

[0089] Several drivers were recruited, and after observing videos on driving simulators, a complex scenario evaluation, or criticality quantification, was required for the test scenarios. The quantification criteria included... Figure 3 As shown, the subjective evaluation matrix focuses on two aspects of the test scenario: the degree to which the driver perceives a collision risk and whether the driver is willing to control the vehicle to improve traffic efficiency; the subjective evaluation matrix is ​​used to determine the subjective evaluation results corresponding to this complex test scenario.

[0090] This invention assigns greater weight to the risk dimension in the evaluation matrix. To ensure the rationality of subjective quantification, this embodiment selects eight different drivers to quantify each test scenario, and uses a weighted average to obtain the subjective quantification result Lab for the nth test scenario. n The specific weighting process is as follows:

[0091]

[0092] In the formula, j represents the scene label value; in this embodiment, four types of labels are set; m j Select the number of people corresponding to the j label for each driver.

[0093] Using only subjective methods for quantifying the criticality of complex test scenarios can lead to a decrease in critical test scenario coverage. Therefore, it is necessary to use objective methods for quantifying the criticality of complex test scenarios to obtain a more comprehensive picture of critical test scenarios.

[0094] (2.2) Critical Quantification Methods for Objective and Complex Test Scenarios

[0095] The key quantification method for objective and complex test scenarios is constructed based on the risk field method. The risk field method is used to construct risk fields for traffic elements and road elements respectively, and to construct field strength attenuation coefficients using meteorological elements. Then, the risk fields generated by the main vehicle and the environment are constructed using the above field strengths respectively, and the interaction between them is used to construct interactive field strength features. The interactive field strength feature (IFF) is defined as the quantification result of the objective method, which integrates the three heterogeneous elements of traffic, road and meteorological elements of the scenario into a single quantification result, thereby objectively describing the driving risk of the main vehicle in the current driving environment.

[0096] like Figure 4 As shown, the calculation process of the interaction field features is as follows:

[0097] First, the road potential field is generated by road elements, traffic elements, and meteorological elements respectively. Traffic situation and adverse weather field intensity degradation factors Based on this, the environmental field strength was constructed. Then, according to The same method is used to calculate the risk field generated by the main vehicle. Finally, discretization is performed based on the Delaunay triangular discretization method. and Given two continuous non-uniform potential fields, calculate the interaction value IFF between the two discrete potential fields. n,t ; and The stronger the repulsion between them, the stronger the corresponding IFF. n,t The larger the value, the greater the safety risk to the vehicle posed by the driving environment at time point t in the nth scenario.

[0098] The traffic potential field The calculation method is as follows:

[0099] Traffic potential field generated by a single vehicle m Defined as a representation of the collision risk between itself and its surroundings based on motion parameters such as position, velocity, and acceleration, its mapping relationship is expressed as follows:

[0100]

[0101] The expression is:

[0102]

[0103] Among them, M m The virtual mass is determined by the type and actual mass of vehicle m; γ R This describes the change in the road surface adhesion coefficient μ caused by rainwater. r c represents the coefficient of friction reduction on the road surface after rainfall. r It is a constant that determines the risk growth rate as it approaches m, c r Set it to 0.7; It is the spatiotemporal distance between any point in the spatiotemporal domain and m, and its construction process takes into account vehicle speed.

[0104] The road potential field The calculation method is as follows:

[0105] Road potential field Two important road elements are mainly considered: one is the potential field generated by static obstacles. Such as obstacles; another type is the potential field generated by the lane. The resulting risks are evenly distributed; this embodiment calculates... When the speed is set to 0, the result is obtained corresponding to a static obstacle. For lanes, different lane lines ε will pose different levels of risk to the driver. For example, when the driver approaches a dashed line and prepares to change lanes, the risk to overcome is much smaller than when approaching a double solid line. This is because driving a vehicle across a double solid line violates traffic regulations, thus leading to greater driving risks. The expression is:

[0106]

[0107] Where, N L Indicates the number of lanes; I i The meaning of the lane line linearity ε for the i-th lane is that the value corresponding to the solid line is greater than the value corresponding to the dashed line; σ is a fixed value, the magnitude of which determines when the vehicle approaches the lane line. growth rate; The expression is:

[0108]

[0109] The intelligent vehicle perception system for weather conditions mainly considers rain and fog, modeling the impact of rain and fog on the vehicle's perception module; and assessing the adverse weather field intensity degradation factors generated by rain and fog on the vehicle's perception module. Affected by meteorological factors The coupling effect of the longitudinal distance Dis between the target and the vehicle is adopted. This indicates that the choice of W depends on the calibration of different sensors; This represents the risk amplification factor for the scenario compared to the case without adverse weather conditions. The modeling process is to determine Dis and The variation pattern of the intensity degradation factor under different combinations.

[0110] This embodiment will The construction method was applied to the YOLO-v5 target recognition function in the camera perception module of an intelligent vehicle under rainy and foggy weather; in order to obtain Dis and The combined parameters are used to construct W by generating images using PreScan and Simulink, such as... Figure 5 As shown. First, different... The system first identifies different Dis values ​​in various scenarios and records the current Dis value and corresponding image using Simulink. Then, it inputs the acquired image into YOLO-v5 to identify targets in the image and records the Dis values. The IOU recognition results under different combinations are defined as IOU. W Again, no record. The IOU recognition results under different Dis are defined as IOU.O Finally, Dis, and perceived attenuation coefficient Θ W Represented by the x-axis, y-axis, and z-axis respectively, such as Figure 6 As shown.

[0111] IOU W and IOU O The absolute difference between them is defined as Θ W Θ was obtained respectively in the same Dis scene under the influence of weather conditions and without the influence of weather conditions. W The expression is:

[0112] Θ W =IOU W -IOU O (8)

[0113] Records included rainfall intensity R a Dis and fog visibility (Vis) Different combinations of IOU W As a result, Θ is constructed. W ; The expression is:

[0114]

[0115] Where, β T This is the object category coefficient, used to determine the degree of risk posed by different target objects to one's own vehicle.

[0116] Based on the three potential field calculation methods described above, the potential field at time t in scenario n is obtained. expression:

[0117]

[0118] for The establishment of its value is determined by and The maximum strength between and multiplied by Confirmed. Figure 7 As shown, different [types of structures] were constructed using the above method. Including R a =15mm / h and Vis=30m The specific results show that The original strength was Magnify, and With Dis and This method is relevant. It can accurately describe the distribution of potential field intensity in an environment.

[0119] In obtaining and Next, the continuous potential field is first discretized using the Delaunay triangle discretization method; then, the IFF is calculated using the feature points of each triangular surface. n,t .

[0120] The Delaunay triangle has advantages such as uniqueness, optimality, and closest proximity. It is used to process... and The corresponding triangular surface is obtained. and By using Risk characteristics and Risk characteristics Quantification separately and IFF between n,t ,like Figure 8 As shown; and The calculation method is the same, with Let's take the i-th example as an example: The coordinates of the three vertices are respectively and Correspondingly The expression is:

[0121]

[0122]

[0123] According to the law of universal gravitation, IFF n,t Depend on and Quantization, its expression is:

[0124]

[0125] in, It refers to the i-th scene in the n-th scene. The coordinates at time t; It refers to the j-th scene in the n-th scene. The coordinates at time t;

[0126] Take the largest IFF from all time points in the scene. n,t As a key quantitative result in objective and complex testing scenarios, IFF n To ensure the consistency of the dimensions of key quantitative results, the IFF will be used. n Normalization is performed.

[0127] (2.3) The final quantitative result C of the scenario is obtained by averaging the key quantitative results of subjective and objective test scenarios.n Its expression is:

[0128]

[0129] (3) Use soft clustering algorithm to quantify the criticality of test scenarios.

[0130] In the actual process of criticality quantification in complex test scenarios, it was found that subjective methods are time-consuming and labor-intensive, and recruiting different drivers to quantify the test scenarios incurs significant human resource costs. Therefore, to improve the efficiency of criticality quantification, it is necessary to use a soft clustering algorithm to quantify the criticality of unquantified test scenarios. The soft clustering algorithm used in this invention considers the imbalance of sample sizes because the probability of critical test scenarios occurring in actual collected scenarios is extremely low. If only traditional clustering algorithms are used, key features may be ignored. The soft clustering algorithm is used to automatically quantify the criticality of other scenarios based on the quantified criticality results, outputting the final criticality quantification results for complex test scenarios.

[0131] Soft clustering algorithms address the strict membership relationships of traditional hard clustering algorithms, where membership values ​​are either 0 or 1. Using fuzzy set theory, they expand the original membership values ​​to any value between 0 and 1. A sample can belong to different clusters with different membership values, thus greatly improving the clustering algorithm's ability to handle real-world datasets. The Generalequalization Fuzzy C-means Clustering Algorithm (GEFCM) improves upon the optimization criteria of the traditional fuzzy C-means clustering algorithm by considering sample size in the objective function design. This effectively alleviates the inaccuracy of clustering results caused by the small proportion of key scenarios in practical applications. The objective function of GEFCM is:

[0132]

[0133] And the following constraints are met:

[0134]

[0135] In the formula, m represents the ambiguity control coefficient; c i x represents the cluster center of the i-th cluster (i = 1, 2, ..., c), where c represents the total number of clusters; j For test scenario data, j = 1, 2, ..., n represents a total of n test scenario data to be clustered; U = (u ij ) c*n Let u be the membership matrix. ij The membership degree is represented by x. j The degree to which a cluster belongs to the i-th cluster; Dis(x)j ,c i ) represents scenario x j With scene c i The similarity between them needs to be considered separately for traffic N in the scenario. T Road N R and meteorology N W The parameter values ​​are specifically defined as follows:

[0136]

[0137] In the formula, This represents the average DTW distance between the vehicle trajectories at all corresponding locations in the two scenarios; a new objective function is constructed based on the GEFCM objective function and constraints using the Lagrange multiplier method:

[0138]

[0139] For u in equation (16) ij and c i Taking the partial derivative and setting the result to 0, we obtain:

[0140]

[0141] By summing i from both sides, we get:

[0142]

[0143] Thus, we obtain the Lagrange multiplier λ. j The expression:

[0144]

[0145] λ j Substituting into the original expression, we get u ij Iterative formula:

[0146]

[0147] Soft clustering algorithms can be used to obtain quantitative results for test scenarios that have not undergone subjective and objective key quantification, such as... Figure 9 As shown. It can be seen that u ij and c i The expression for u is a series of coupled equations, and an analytical solution cannot be obtained. Therefore, u is obtained through iteration. ij and c i The estimated values ​​and the specific iterative solution process are shown in the table below:

[0148]

[0149]

Claims

1. A key quantitative method integrating subjective and objective methods for complex testing scenarios of intelligent vehicles, characterized in that, Includes the following steps: (1) Processing complex test scenario datasets: The data at time t of the nth complex test scenario is split into three scenario elements, including meteorological elements. Transportation elements and road elements (2) Key quantitative method integrating subjective and objective factors (2.1) Key Quantification Methods for Subjective and Complex Test Scenarios First, the three scene elements mentioned above are reproduced using virtual simulation software, that is, a complex scene identical to the real environment is constructed in the virtual simulation scene. Then the driving perspective of the main vehicle is displayed on the driving simulator; Several drivers were recruited, and after observing videos on a driving simulator, a complex scenario evaluation, i.e., criticality quantification, was required for the test scenario. The subjective evaluation matrix focused on two aspects of the test scenario: the degree to which the driver believed there was a collision risk in the scenario and whether the driver was willing to control the vehicle to improve traffic efficiency. The subjective evaluation matrix was used to determine the subjective evaluation results corresponding to the complex test scenario. To ensure the reasonableness of subjective quantification, several different drivers were selected to quantify each test scenario, and the subjective quantification result Lab for the nth test scenario was finally obtained by weighting the results by the number of drivers. n The specific weighting process is as follows: In the formula, j represents the scene label value; m j Select the number of drivers corresponding to the label j; (2.2) Critical Quantification Methods for Objective and Complex Test Scenarios The key quantification method for objective and complex test scenarios is constructed based on the risk field method. The risk field method is used to construct risk fields for traffic elements and road elements respectively, and to construct field strength attenuation coefficients using meteorological elements. Then, the risk fields generated by the main vehicle and the environment are constructed using the above field strengths respectively, and the interaction between them is used to construct interactive field strength features. The interactive field strength features (IFF) are defined as the quantification result of the objective method, which integrates the three heterogeneous elements of traffic, road and meteorological elements of the scenario into a single quantification result, thereby objectively describing the driving risk of the main vehicle in the current driving environment. The calculation process for the interactive field features is as follows: First, the road potential field is generated by road elements, traffic elements, and meteorological elements respectively. Traffic situation and adverse weather field intensity degradation factors Based on this, the environmental field strength was constructed. Then, according to The same method is used to calculate the risk field generated by the main vehicle. Finally, discretization is performed based on the Delaunay triangular discretization method. and Given two continuous non-uniform potential fields, calculate the interaction value IFF between the two discrete potential fields. n,t ; and The stronger the repulsion between them, the stronger the corresponding IFF. n,t The larger the value, the greater the safety risk to the vehicle posed by the driving environment at time point t in the nth scenario; Take the largest IFF from all time points in the scene. n,t As a key quantitative result in objective and complex testing scenarios, IFF n To ensure the consistency of the dimensions of key quantitative results, the IFF will be used. n Perform normalization processing; (2.3) The final quantitative result C of the scenario is obtained by averaging the key quantitative results of subjective and objective test scenarios. n Its expression is: (3) Use soft clustering algorithm to quantify the criticality of test scenarios. The soft clustering algorithm is used to perform criticality quantification on unquantized test scenarios. The soft clustering algorithm takes into account the imbalance of sample numbers. Based on the quantified criticality results, the soft clustering algorithm is used to automatically quantify the criticality of other scenarios and output the final criticality quantification results of complex test scenarios.

2. The key quantitative method for subjective and objective fusion in complex testing scenarios for intelligent vehicles according to claim 1, characterized in that: In step (1), meteorological elements Including rainfall intensity R a Fog visibility (Vis); Traffic factors This includes the vehicle's lateral and longitudinal position, speed, and acceleration; road elements. This includes the road surface adhesion coefficient μ, lane position, and lane line linearity.

3. The key quantitative method for subjective and objective fusion in complex testing scenarios for intelligent vehicles according to claim 1, characterized in that: In step (2.2), the traffic potential field The calculation method is as follows: Traffic potential field generated by a single vehicle m Defined as a representation of the collision risk with its surroundings based on motion parameters of position, velocity, and acceleration, its mapping relationship is expressed as follows: The expression is: Among them, M m The virtual mass is determined by the type and actual mass of vehicle m; γ R This describes the change in the road surface adhesion coefficient μ caused by rainwater. r c represents the coefficient of friction reduction on the road surface after rainfall. r It is a constant that determines the risk growth rate as it approaches m, c r Set it to 0.7; It is the spatiotemporal distance between any point in the spatiotemporal domain and m, and its construction process takes into account vehicle speed; The road potential field The calculation method is as follows: Road potential field Consider two important types of road elements: one is the potential field generated by static obstacles. Another type is the potential field generated by the lane. The resulting risks are uniformly distributed, in calculation When the speed is set to 0, the result is obtained corresponding to a static obstacle. For a lane, the potential field generated by the lane The expression is: Where, N L Indicates the number of lanes; I i The meaning of the lane line linearity ε for the i-th lane is that the value corresponding to the solid line is greater than the value corresponding to the dashed line; σ is a fixed value, the magnitude of which determines when the vehicle approaches the lane line. growth rate; The expression is: The intelligent vehicle perception system considers both rain and fog conditions, modeling the impact of rain and fog on the vehicle's perception module; and assessing the adverse weather field intensity degradation factors generated by rain and fog conditions on the vehicle's perception module. by The coupling effect of the longitudinal distance Dis between the target and the vehicle is adopted. This indicates that the choice of W() depends on the calibration of different sensors; This represents the risk amplification factor for the scenario compared to the case without adverse weather conditions. The modeling process is to determine Dis and The variation pattern of the intensity degradation factor under different combinations; Based on the three potential field calculation methods described above, the potential field at time t in scenario n is obtained. expression: for The establishment of its value is determined by and The maximum strength between and multiplied by Sure.

4. The key quantitative method for subjective and objective fusion in complex testing scenarios for intelligent vehicles according to claim 3, characterized in that: Will The construction method was applied to the YOLO-v5 target recognition function in the camera perception module of an intelligent vehicle under rainy and foggy weather; in order to obtain Dis and The comprehensive parameter combination is used to construct W() by generating images using PreScan and Simulink; firstly, different parameters are constructed in PreScan. The system first identifies different Dis values ​​in various scenarios and records the current Dis value and corresponding image using Simulink. Then, it inputs the acquired image into YOLO-v5 to identify targets in the image and records the Dis values. The IOU recognition results under different combinations are defined as IOU. W Again, no record. The IOU recognition results under different Dis are defined as IOU. O Finally, Dis, and perceived attenuation coefficient Θ W Represented by the x-axis, y-axis, and z-axis respectively; IOU W and IOU O The absolute difference between them is defined as Θ W Θ was obtained respectively in the same Dis scene under the influence of weather conditions and without the influence of weather conditions. W The expression is: I W =|IOU W -IOU O | Records included rainfall intensity R a Dis and fog visibility (Vis) Different combinations of IOU W As a result, Θ is constructed. W ; The expression is: Where, β T This is the object category coefficient, used to determine the degree of risk posed by different target objects to one's own vehicle.

5. The key quantitative method for subjective and objective fusion in complex testing scenarios for intelligent vehicles according to claim 1, characterized in that: In step (2.2), the interaction value IFF between the two discrete potential fields n,t The calculation method is as follows: In obtaining and Next, the continuous potential field is first discretized using the Delaunay triangle discretization method; then, the IFF is calculated using the feature points of each triangular surface. n,t ; The Delaunay triangle possesses the advantages of uniqueness, optimality, and closest proximity. It is used to process... and The corresponding triangular surface is obtained. and By using Risk characteristics and Risk characteristics Quantification separately and IFF between n,t ; and The calculation method is the same, with Let's take the i-th example as an example: The coordinates of the three vertices are respectively and Correspondingly The expression is: According to the law of universal gravitation, IFF n,t Depend on and Quantization, its expression is: in, It refers to the i-th scene in the n-th scene. The coordinates at time t; It refers to the j-th scene in the n-th scene. The coordinates at time t.

6. The key quantitative method for subjective and objective fusion in complex testing scenarios for intelligent vehicles according to claim 1, characterized in that: In step (3), the soft clustering algorithm addresses the strict membership relationship of traditional hard clustering algorithms, where the membership degree value is either 0 or 1. It uses fuzzy set theory to expand the original membership degree to any value between 0 and 1, thereby improving the clustering algorithm's ability to process real-world datasets. The General Equilibrium Fuzzy C-means Clustering Algorithm (GEFCM) improves upon the optimization criteria of the traditional fuzzy C-means clustering algorithm. The objective function of GEFCM is: And the constraints are satisfied: In the formula, m represents the ambiguity control coefficient; c i x represents the cluster center of the i-th cluster, i = 1, 2, ..., c, where c represents the total number of clusters; j For test scenario data, j = 1, 2, ..., n represents a total of n test scenario data to be clustered; U = (u ij ) c*n Let u be the membership matrix. ij The membership degree is represented by x. j The degree to which a cluster belongs to the i-th cluster; Dis(x) j ,c i ) represents scenario x j With scene c i The similarity between them needs to be considered separately for traffic N in the scenario. T Road N R and meteorology N W The parameter values ​​are specifically defined as follows: In the formula, This represents the average DTW distance between the vehicle trajectories at all corresponding locations in the two scenarios; a new objective function is constructed based on the GEFCM objective function and constraints using the Lagrange multiplier method: For the objective function formula of GEFCM, u ij and c i Taking the partial derivative and setting the result to 0, we obtain: By summing i from both sides, we get: Thus, we obtain the Lagrange multiplier λ. j The expression: λ j Substituting into the original expression, we get u ij Iterative formula: Soft clustering algorithms can be used to obtain quantitative results for test scenarios that have not undergone subjective and objective key quantification. ij and c i The expression for u is a series of coupled equations, and an analytical solution cannot be obtained. Therefore, u is obtained through iteration. ij and c i The estimated value.

Images