A method for extracting function scenarios applicable to high-speed autonomous driving

Abstract

A method for extracting high-speed autonomous driving function scenarios, the method comprising: fixing a data acquisition device on the vehicle itself and collecting data related to the vehicle itself and target vehicles in real time; creating a data partition serial number column based on the vehicle speed of the vehicle itself and the number of surrounding target vehicles, so that the collected data is partitioned according to the data partition serial number column of the vehicle speed of the vehicle itself and the number of surrounding target vehicles; performing data cleaning on each obtained partition; based on the cleaned data, processing the driving parameters of the vehicle itself to obtain the predicted driving area of the vehicle itself, and marking the objects entering the predicted driving area of the vehicle itself as front objects; classifying high-speed autonomous driving function scenarios, and extracting data of following scenarios, cut-in scenarios and cut-out scenarios of high-speed autonomous driving function scenarios; after integrating the extracted scenario data, randomly select non-repeating n1 and n2 integers between 1 and N respectively, and the corresponding values are the numbers for verification, and verify the extracted data according to the numbers.

Description

Technical Field

[0001] The present invention relates to the technical field of high-speed autonomous driving function scenarios, and particularly relates to a method for extracting high-speed autonomous driving function scenarios Background Art

[0002] As domestic intelligent vehicles gradually reach L3 / L4 levels, mainly involving ADAS functions such as congestion following, highway valet driving, and valet parking; among them, the most commonly used is the high-speed autonomous driving function, which can automatically adjust the following speed according to the speed of the vehicle ahead and the lane lines, and assist the driver to keep the vehicle within its own lane.

[0003] Extract representative scenarios related to the high-speed autonomous driving function, perform data slicing and extraction, and analyze the data segments to provide corresponding scenario data support for the research and development and experimental verification of the high-speed autonomous driving function decision model.

[0004] However, the extremely high data collection frequency and the extremely long data collection mileage have caused the vehicle data to grow exponentially and explosively, with the number of included scenarios reaching millions. How to extract representative and effective scenarios for the high-speed autonomous driving function from them for data extraction has become a major difficulty.

[0005] In the prior art, the patent document CN114064656A discloses "a method for identifying and transforming autonomous driving scenarios based on a roadside perception system", which can identify and transform a large amount of driving data collected by the roadside perception system and output sliced data with function tags to provide data sources for the construction of a large-scale database for autonomous driving training. In the case of relying on the sensors of the host vehicle and only having one host vehicle, the present invention accurately and quickly slices the scenarios. The patent document CN114067243A discloses "a method, system, device, and medium for identifying autonomous driving scenarios", which can accurately distinguish the current environmental characteristics and differences, realize in-depth clustering of the environment, and assist in the migration of autonomous driving decision-making algorithms, rather than the scenario slicing method of the present invention.

[0006] Therefore, the existing technical means cannot extract representative and effective scenario data from a large amount of data of the high-speed autonomous driving function. Summary of the Invention

[0007] The present invention solves the problem that the existing technical means cannot extract representative and effective scenario data from a large amount of data of the high-speed autonomous driving function.

[0008] A method for extracting high-speed autonomous driving function scenarios according to the present invention includes the following steps:

[0009] Step S1, fix the data collection device on the vehicle itself and collect relevant data of the vehicle itself and the target vehicle;

[0010] Step S2: Create a data partition serial number column based on the vehicle's own speed and the number of surrounding target vehicles, and partition the collected data according to the data partition serial number column of the vehicle's own speed and the number of surrounding target vehicles;

[0011] Step S3: Perform data cleaning on each partition obtained in Step S2;

[0012] Step S4: Based on the data cleaned in Step S3, process the driving parameters of the vehicle to obtain the predicted driving area of the vehicle, and mark the objects entering the predicted driving area of the vehicle as front targets;

[0013] Step S5: Classify the high-speed automatic driving function scenarios, and extract the data of the following vehicle scenarios, cut-in scenarios and cut-out scenarios in the high-speed automatic driving function scenarios;

[0014] Step S6: After integrating the extracted scenario data, randomly select non-repeating n1 and n2 integers between 1 and N respectively, and the corresponding values are the numbers of the tests. Check the extracted data according to the numbers.

[0015] Further, in an embodiment of the present invention, in Step S1, the data acquisition device includes a millimeter-wave radar, an intelligent camera, and a lidar.

[0016] Further, in an embodiment of the present invention, in Step S1, the vehicle-related data includes the vehicle's own speed, the vehicle's lateral acceleration, the vehicle's longitudinal acceleration, the vehicle's steering wheel angle, the vehicle's heading angle, the vehicle's longitude, and the vehicle's latitude;

[0017] The target vehicle-related data includes the relative longitudinal distance between the target vehicle and the vehicle, the relative lateral distance between the target vehicle and the vehicle, the relative speed between the target vehicle and the vehicle, the absolute speed of the target vehicle, the absolute lateral acceleration of the target vehicle, and the absolute longitudinal acceleration of the target vehicle.

[0018] Further, in an embodiment of the present invention, in Step S3, performing data cleaning on each partition includes the following steps:

[0019] Step S301: Perform second-order spline curve interpolation upsampling on the vehicle's own speed, the relative speed between the target vehicle and the vehicle, and the absolute speed of the target vehicle;

[0020] Step S302: Construct the standard deviation of the Gaussian kernel function, determine the mean value, and then start performing convolution operations;

[0021] Step S303: Downsample the vehicle's own speed, the relative speed between the target vehicle and the vehicle, and the absolute speed of the target vehicle at equal intervals to their original sampling frequencies.

[0022] Further, in an embodiment of the present invention, in the step S303, the convolution operation formula is as follows:

[0023] In the formula, y(n) is the updated value after filtering, G(i) is the Gaussian kernel function template, and h(n - i) is the original data of the vehicle speed of the host vehicle, the relative speed between the target vehicle and the host vehicle, and the absolute speed between the target vehicle and the host vehicle.

[0024] Further, in an embodiment of the present invention, in the step S4, obtaining the predicted driving area of the host vehicle after processing the driving parameters of the host vehicle includes the following steps:

[0025] Step S401, obtaining the driving curvature of the host vehicle and the steering wheel angle data of the host vehicle that have been collected;

[0026] Step S402, calculating the weighted average of the driving curvature of the host vehicle and the steering wheel angle data of the host vehicle, taking the driving trajectory of the host vehicle as the origin, and extending the forward direction of the host vehicle as the predicted path of the host vehicle;

[0027] Step S403, translating the predicted path of the host vehicle to both sides along the radial direction by 1 / 2 lane width to obtain the predicted driving area of the host vehicle.

[0028] Further, in an embodiment of the present invention, in the step S5, the following car - following scenarios include car - following start, uniform car - following, car - following acceleration, car - following deceleration, car - following stop, overtaking in the same lane, and curve driving.

[0029] Further, in an embodiment of the present invention, in the step S5, the principle of data extraction for the car - following scenario, cut - in scenario, and cut - out scenario of the high - speed autonomous driving function scenario is as follows:

[0030] Taking the relative speed between the target vehicle and the host vehicle as the extraction condition, when the relative speed is between different intervals, the target vehicle and the host vehicle are in different interaction states;

[0031] Creating an identification column for different interaction states of the target vehicle and the host vehicle. When the relative speed > 0, the identification column of this column is 0, otherwise it is 1;

[0032] Taking the derivative of the identification column for different interaction states of the target vehicle and the host vehicle to obtain the derivative column of its identification column, and setting the end value to 1;

[0033] Taking the reverse derivative of the identification column for different interaction states of the target vehicle and the host vehicle to obtain the reverse derivative column of its identification column, and setting the first value to 1;

[0034] Screen the time points corresponding to the non-zero values in the derivative sequence and the anti-derivative sequence. Then, the start and end times of each relative speed interval are composed of the time axis corresponding to the non-zero points in the anti-derivative sequence and the zero points in the derivative sequence.

[0035] The relative speed intervals in odd positions form the vehicle state corresponding to the first interval, and those in even positions are the opposite.

[0036] When the interval between two consecutive identical states is less than 1 s, they are merged.

[0037] The initial actions of the target vehicle and the host vehicle are both line following.

[0038] Further, in an embodiment of the present invention, in step S6, the integration of the extracted scenario data includes the following steps:

[0039] Step S601: Select 150 - 180 extracted scenario data.

[0040] Step S602: Number the scenario data to be tested incrementally starting from 1.

[0041] Step S603: Set the number of scenario data, acceptance limit, and rejection limit for the corresponding test rounds according to the selected amount of scenario data and the expected accuracy rate.

[0042] Further, in an embodiment of the present invention, in step S6, the inspection of the extracted data according to the number is specifically as follows:

[0043] If the number of unqualified data d1 found in the first inspection of the scenario data is less than or equal to Ac1, the inspection result is considered to be accepted.

[0044] If the number of unqualified data d1 found in the first inspection of the scenario data is greater than or equal to Re1, the inspection result is considered to be rejected.

[0045] If the number of unqualified data d1 found in the first inspection of the scenario data is between Ac1 and Re1, a second inspection is carried out according to the plan, and the total number of unqualified data in the two inspections is accumulated, that is, d1 + d2.

[0046] If the sum of d1 and d2 is less than or equal to the acceptance limit Ac2, the inspection result is considered to be accepted.

[0047] If the sum of d1 and d2 is greater than or equal to the rejection limit Re2, the inspection result is considered to be rejected.

[0048] The present invention solves the problem that the existing technical means cannot extract representative and effective scenario data from the large amount of data of high-speed autonomous driving functions. The specific beneficial effects include:

[0049] 1. A method for extracting high-speed autonomous driving function scenarios according to the present invention innovatively uses a partitioning algorithm, a cleaning algorithm, a scenario recognition algorithm based on relative speed, and a scenario extraction algorithm for objective data such as the speed of the vehicle itself, the speed of the target vehicle, and the longitudinal and lateral distances between the target vehicle and the vehicle itself in natural driving scenario data, so as to achieve efficient and automated extraction of high-speed autonomous driving function scenario data from big data.

[0050] 2. A method for extracting high-speed autonomous driving function scenarios according to the present invention can more accurately correct interfering function scenario data through a data cleaning algorithm based on a Gaussian kernel function.

[0051] 3. A method for extracting high-speed autonomous driving function scenarios according to the present invention uses a scenario recognition algorithm based on the relative speed of the target vehicle to reduce algorithm parameters.

[0052] 4. A method for extracting high-speed autonomous driving function scenarios according to the present invention creates a data partition serial number column based on the vehicle speed and the number of surrounding target objects, so that the collected data is partitioned according to this column, and a partitioning algorithm is used for parallel calculation, greatly shortening the calculation time. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] The above and / or additional aspects and advantages of the present invention will become apparent and be readily understood from the following description of the embodiments in conjunction with the accompanying drawings, where:

[0054] Figure 1 is the overall flowchart of the function scenario extraction method described in the specific implementation manner.

[0055] Figure 2 is the partitioning flowchart described in the specific implementation manner.

[0056] Figure 3 is the quadratic spline solution diagram described in the specific implementation manner.

[0057] Figure 4 is the schematic diagram of the predicted driving area described in the specific implementation manner.

[0058] Figure 5 is the key following scenario diagram described in the specific implementation manner.

[0059] Figure 6 is the vehicle state extraction method diagram described in the specific implementation manner.

[0060] Figure 7 is the key cutting-in scenario diagram described in the specific implementation manner.

[0061] Figure 8 is the key cutting-out scenario diagram described in the specific implementation manner.

[0062] Figure 9 is the secondary sampling inspection flowchart described in the specific implementation manner. Specific implementation manner

[0063] The following will clearly and completely describe various implementation manners of the present invention in conjunction with the accompanying drawings. The embodiments described by referring to the accompanying drawings are exemplary and are intended to explain the present invention, and should not be construed as a limitation to the present invention.

[0064] A method for extracting high-speed autonomous driving function scenarios described in this implementation manner includes the following steps:

[0065] Step S1: Fix the data acquisition device on the vehicle itself and collect relevant data of the vehicle itself and the target vehicle.

[0066] Step S2: Create a data partition serial number column based on the vehicle speed of the vehicle itself and the number of surrounding target vehicles, and partition the collected data according to the data partition serial number column of the vehicle speed of the vehicle itself and the number of surrounding target vehicles.

[0067] Step S3: Perform data cleaning on each partition obtained in Step S2.

[0068] Step S4: Based on the data cleaned in Step S3, process the driving parameters of the vehicle itself to obtain the predicted driving area of the vehicle itself, and mark the target objects entering the predicted driving area of the vehicle itself as front target objects.

[0069] Step S5: Classify the high-speed autonomous driving function scenarios, and extract data of the following vehicle scenarios, cut-in scenarios, and cut-out scenarios of the high-speed autonomous driving function scenarios.

[0070] Step S6: After integrating the extracted scenario data, randomly select non-repeating n1 and n2 integers between 1 and N respectively, and the corresponding values are the numbers of the inspection. Inspect the extracted data according to the numbers.

[0071] In this implementation manner, in Step S1, the data acquisition device includes a millimeter-wave radar, an intelligent camera, and a lidar.

[0072] In this implementation manner, in Step S1, the relevant data of the vehicle itself includes the vehicle speed of the vehicle itself, the lateral acceleration of the vehicle itself, the longitudinal acceleration of the vehicle itself, the steering wheel angle of the vehicle itself, the heading angle of the vehicle itself, the longitude of the vehicle itself, and the latitude of the vehicle itself.

[0073] The relevant data of the target vehicle includes the relative longitudinal distance between the target vehicle and the vehicle itself, the relative lateral distance between the target vehicle and the vehicle itself, the relative speed between the target vehicle and the vehicle itself, the absolute speed of the target vehicle, the absolute lateral acceleration of the target vehicle, and the absolute longitudinal acceleration of the target vehicle.

[0074] In this embodiment, in step S3, each partition performs data cleaning, including the following steps:

[0075] Step S301, perform second-order spline curve interpolation upsampling on the vehicle speed of the host vehicle, the relative speed between the target vehicle and the host vehicle, and the absolute speed between the target vehicle and the host vehicle;

[0076] Step S302, construct the standard deviation of the Gaussian kernel function, determine the mean value, and then start performing convolution operations;

[0077] Step S303, evenly downsample the vehicle speed of the host vehicle, the relative speed between the target vehicle and the host vehicle, and the absolute speed between the target vehicle and the host vehicle to their original sampling frequencies.

[0078] In this embodiment, in step S303, the convolution operation formula is:

[0079] In the formula, y(n) is the updated value after filtering, G(i) is the Gaussian kernel function template, and h(n - i) is the original data of the vehicle speed of the host vehicle, the relative speed between the target vehicle and the host vehicle, and the absolute speed between the target vehicle and the host vehicle.

[0080] In this embodiment, in step S4, obtaining the predicted driving area of the host vehicle after processing the driving parameters of the host vehicle includes the following steps:

[0081] Step S401, obtain the driving curvature of the host vehicle and the data of the steering wheel angle of the host vehicle that have been collected;

[0082] Step S402, calculate the weighted average of the driving curvature of the host vehicle and the data of the steering wheel angle of the host vehicle, and take the driving trajectory of the host vehicle as the origin, and extend the forward direction of the host vehicle as the predicted path of the host vehicle;

[0083] Step S403, translate the predicted path of the host vehicle 1 / 2 lane width along the radial direction to both sides respectively to obtain the predicted driving area of the host vehicle.

[0084] In this embodiment, in step S5, the following following scenarios include following start, uniform following, following acceleration, following deceleration, following stop, overtaking in the same lane, and driving in a curve.

[0085] In this embodiment, in step S5, the principle for extracting data from the following scenarios, cut-in scenarios, and cut-out scenarios of the high-speed automatic driving function scenarios is:

[0086] Taking the relative speed between the target vehicle and the host vehicle as the extraction condition, when the relative speed is between different intervals, the target vehicle and the host vehicle are in different interaction states;

[0087] Create an identification column for different interaction states between the target vehicle and the host vehicle. When the relative speed > 0, the value of this column is 0; otherwise, it is 1.

[0088] Derive the identification column for different interaction states between the target vehicle and the host vehicle to obtain its derivative column, and set the last value to 1.

[0089] Derive the identification column for different interaction states between the target vehicle and the host vehicle in the reverse direction to obtain its anti-derivative column, and set the first value to 1.

[0090] Filter the time points corresponding to the non-zero values in the derivative column and the anti-derivative column. Then, the time axis corresponding to the non-zero points in the anti-derivative column and the zero points in the derivative column forms the start and end times of each relative speed interval.

[0091] The relative speed interval of the odd order corresponds to the state of the host vehicle in the first interval, and the even order is the opposite.

[0092] When the interval between two consecutive identical states is less than 1 s, merge them.

[0093] The initial actions of both the target vehicle and the host vehicle are to follow the line.

[0094] In this embodiment, in step S6, the integration of the extracted scenario data includes the following steps:

[0095] Step S601: Select 150 - 180 extracted scenario data.

[0096] Step S602: Number the scenario data to be tested incrementally starting from 1.

[0097] Step S603: Set the number of scenario data, acceptance limit, and rejection limit for the corresponding test rounds according to the selected amount of scenario data and the expected accuracy rate.

[0098] In this embodiment, in step S6, the inspection of the extracted data according to the numbers is specifically as follows:

[0099] If the number of unqualified data d1 found in the first inspection of the scenario data is less than or equal to Ac1, the inspection result is considered to be accepted.

[0100] If the number of unqualified data d1 found in the first inspection of the scenario data is greater than or equal to Re1, the inspection result is considered to be rejected.

[0101] If the number of unqualified data d1 found in the first inspection of the scenario data is between Ac1 and Re1, conduct a second inspection according to the plan and accumulate the number of unqualified data in the two inspections, that is, d1 + d2.

[0102] If the sum of d1 and d2 is less than or equal to the acceptance limit Ac2, the inspection result is considered to be accepted.

[0103] If the sum of d1 and d2 is greater than or equal to the rejection limit Re2, the inspection result is considered to be rejected.

[0104] This embodiment is based on a method for extracting high-speed autonomous driving function scenarios described in the present invention, combined with Figure 1 This embodiment can be better understood, and a practical embodiment is provided in combination with a specific object:

[0105] S1 Data acquisition:

[0106] S101: Data acquisition includes millimeter-wave radar, intelligent cameras, and lidar fixed around the scenario acquisition vehicle to collect objective data of surrounding traffic participants in real time. The collected objective data can meet the data requirements for scenario extraction, and the included content is as follows:

[0107] This vehicle: This vehicle's speed, lateral acceleration, longitudinal acceleration, steering wheel angle, heading angle, longitude, and latitude of this vehicle.

[0108] Target vehicle: Relative longitudinal distance from this vehicle, relative lateral distance from this vehicle, relative speed from this vehicle, absolute speed of the target vehicle, absolute lateral acceleration of the target vehicle, and absolute longitudinal acceleration of the target vehicle.

[0109] S2 Create a data partition serial number column based on the vehicle speed and the number of surrounding targets, so that the collected data is partitioned according to this column, ensuring that the number of data rows in each partition is roughly the same. The partitioning rules are as follows, and the process is shown in Figure 2 :

[0110] S201 First, select the data where the vehicle speed of this vehicle is not 0 among all the data, and take the data within continuous time points as a partition;

[0111] S202 Merge the partitions where the difference between the end time of each partition in a and the start time of the next partition is less than 1 s;

[0112] S203 Set the partition end threshold i to 10;

[0113] S204 Calculate the difference between the number of rows in each partition and the maximum number of rows in the partition. If it is greater than i%, then re-cut according to the condition that the number of surrounding target data in each partition is less than 1, that is, re-cut within each interval where the vehicle speed is not 0;

[0114] S205 Judge again the difference between the number of rows in each partition and the maximum number of rows in the partition. If it is still greater than i%, then re-partition according to the condition that the number of surrounding vehicles is greater than 1 and the relative distance and relative speed remain stable within 2 s;

[0115] S206 again determines the difference between the number of rows in each partition and the maximum number of rows in the partition. If it is still greater than i%, the value of i is updated to 10*number of loops.

[0116] S207 repeats steps S204 to S206 until the conditions are met;

[0117] If the final i value exceeds 30, the data needs to be cleaned, filtered, and partitioned again to ensure that the difference between the number of data rows in each partition and the maximum number of partition rows is less than 30%.

[0118] S3 data cleaning:

[0119] S301 performs data cleaning steps for each partition;

[0120] S302 performs second-order spline curve interpolation upsampling on the vehicle speed, target vehicle absolute speed, and relative speed data, increasing the data sampling rate to at least 100 Hz. The interpolation method is as follows:

[0121] Assume there are 4 points, x0, x1, x2, x3, and 3 intervals. We need 3 quadratic splines, each of which is ax^2+bx+c, so there are 9 unknowns in total. Figure 3 shown.

[0122] In the figure, there is a quadratic function passing through both endpoints x0 and x3, and two equations can be determined.

[0123] There are two quadratic functions passing through the two intermediate points X1 and x2, and four equations can be determined.

[0124] The middle point must be connected, and the first-order derivatives of the left and right quadratic functions must be equal, that is:

[0125] 2a1x1+b1=2a2x2+b2

[0126] 2a2x2+b2=2a3x3+b3

[0127] Two equations can be determined, and now there are 8 equations.

[0128] Using free boundaries for constraints, 2a0x0+b0=0.

[0129] There are a total of 9 unknowns, and 9 equations are solved simultaneously to complete the upsampling.

[0130] S303 constructs a Gaussian kernel function with a standard deviation of σ and a mean of μ=15, and a Gaussian template truncated with a length of μ±nσ=4 standard deviations outside the mean, as follows;

[0131]

[0132] S304 performs a convolution operation;

[0133] Where y(n) is the updated value after filtering, G(i) is the Gaussian kernel function template, and h(n - i) is the original data.

[0134] S305 performs downsampling, downsampling the data at equal intervals to the original sampling frequency.

[0135] S4 Front target vehicle recognition:

[0136] S401 Obtain the driving curvature of the vehicle collected by the sensor and obtain the steering wheel angle of the vehicle;

[0137] S402 Based on the weighted average of the steering wheel angle and the driving curvature of the vehicle, extend the current driving trajectory of the vehicle with the front bumper center as the origin and the vehicle's forward direction as the prediction path;

[0138] S403 Translate the prediction path radially by 1 / 2 lane width to both sides respectively to obtain the predicted driving area, as Figure 4 shown.

[0139] S404 When the detected target object invades the predicted driving area of the vehicle, then the target object is marked as the front target object.

[0140] S5 Functional scenario recognition:

[0141] S501 Classify the high-speed autonomous driving functional scenarios. How to classify, define the situation where there is a target vehicle in front of the vehicle as a following scenario, and extract the key scenarios in the following scenario. The key scenarios are as Figure 5 shown.

[0142] The principle of the scenario extraction method is as follows:

[0143] Take the relative speed between the vehicle and the front target vehicle as the main extraction condition. When the relative speed is in different intervals, define that the vehicle and the vehicle in front are in different interaction states. For example, when the relative speed is less than 0 km / h, it is the state where the vehicle chases the vehicle in front. When the relative speed is greater than 0 km / h, it is the state where the vehicle approaches the vehicle in front. The extraction methods for different states are as follows, as Figure 6 shown;

[0144] Create an identification column named. When the relative speed > 0, this column is 0, otherwise it is 1;

[0145] Take the derivative of the identification column to obtain the derivative column of the identification column, and set the last value to 1;

[0146] Take the reverse derivative of the identification column to obtain the reverse derivative column of the identification column, and set the first value to 1;

[0147] Screen the time points corresponding to the non-zero values in the derivative sequence and the anti-derivative sequence. Then, the time axis corresponding to the non-zero points in the anti-derivative sequence and the zero points in the derivative sequence forms the start and end times of each relative speed interval.

[0148] The relative speed intervals in odd positions form the vehicle state corresponding to the first interval, and those in even positions are the opposite.

[0149] When the interval between two consecutive identical states is less than 1 s, they are merged.

[0150] The initial actions of both the host vehicle and the target vehicle are following the line.

[0151] Method for extracting starting when following a vehicle:

[0152] Screen the intervals where the relative speed is greater than 0 km / h, the duration is greater than 1 s, the minimum value of the host vehicle speed in this interval is less than 1 km / h, the maximum value of the host vehicle speed is greater than 5 km / h, the average value of the longitudinal acceleration of the host vehicle is greater than 0 m / s 2 , the maximum value of the absolute speed of the target vehicle is less than 160 km / h, the minimum value of the absolute speed of the target vehicle is less than 2 km / h, and the average value of the longitudinal absolute acceleration of the target vehicle is greater than 0 m / s 2 , and the minimum value of the relative distance is less than 10 m.

[0153] Method for extracting constant-speed following:

[0154] Extract the intervals where the relative speed is between -5 and 5 km / h, the duration after merging is greater than 5 s, the minimum value of the host vehicle speed is greater than 5 km / h, the difference between the maximum and minimum values of the host vehicle speed is less than 5 km / h, the maximum absolute value of the steering wheel angle of the host vehicle is less than 30°, and the absolute speed of the target vehicle is less than 160 km / h.

[0155] Method for extracting acceleration when following a vehicle:

[0156] Extract the intervals where the relative speed is greater than 0 km / h, the duration after merging is greater than 1 s; the minimum value of the host vehicle speed is greater than 5 km / h; the absolute value of the steering wheel angle of the host vehicle is less than 30°; the average value of the longitudinal acceleration of the host vehicle is greater than 0 m / s 2 ; the average value of the longitudinal absolute acceleration of the target vehicle is greater than 0 m / s 2 ; the maximum value of the absolute speed of the target vehicle is less than 160 km / h; the minimum value of the absolute speed of the target vehicle is greater than 0 km / h.

[0157] Method for extracting deceleration when following a vehicle:

[0158] Extract the relative speed less than 0 km / h and the duration after merging is greater than 1 s; the minimum value of the vehicle speed is greater than 5 km / h; the maximum value of the absolute value of the vehicle's steering wheel angle is less than 30°; the average value of the vehicle's longitudinal acceleration is less than 0 m / s²; the relative speed is less than -5 km / h; the average value of the relative speed is greater than -60 km / h; the maximum value of the target vehicle's absolute speed is less than 160 km / h; the minimum value of the target vehicle's absolute speed is greater than 0 km / h.

[0159] Following vehicle braking extraction method:

[0160] Extract the relative speed less than 0 km / h and the duration after merging is greater than 1 s; the minimum value of the vehicle speed is less than 1 km / h; the maximum value of the vehicle speed is greater than 5 km / h; the maximum value of the absolute value of the vehicle's steering wheel angle is less than 30°; the average value of the vehicle's longitudinal acceleration is less than 0 m / s 2 ; the maximum value of the target vehicle's absolute speed is less than 160 km / h; the minimum value of the relative distance is less than 15 m; the average value of the target vehicle's longitudinal absolute acceleration is less than 0 m / s 2 .

[0161] Overtaking extraction rule in this lane:

[0162] The relative speed between the vehicle and the vehicle on the immediate right or immediate left is greater than 0 km / h; the maximum value of the absolute value of the vehicle's steering wheel angle is less than 30°; the maximum value of the target vehicle's absolute speed is less than 160 km / h.

[0163] Cornering extraction rule:

[0164] Extract the interval where the radius of the vehicle's driving trajectory is less than 1 km; when the interval between two consecutive intervals is less than 1 s, they are merged.

[0165] S502 data extraction method, define the cut-in critical scenario as follows:

[0166] Definition: In a continuous time period, the overlap rate between the target vehicle and the vehicle is greater than 30%, and the lateral speed of the target vehicle is greater than the set threshold, and the target vehicle is initially on the right front side or the right side, and the final overlap rate is greater than 90%.

[0167] Start time: The overlap rate between the target object and the vehicle just starts to be greater than 30% (adjustable).

[0168] End time: The overlap rate between the target object and the vehicle reaches more than 90% or the distance between the center of the target vehicle and the driving center of the vehicle is less than 0.5 m (adjustable) and the lateral speed is lower than 0.3 m / s (adjustable).

[0169] Initial moment: Push back the start time to the moment when the target object appears or the moment when the target object approaches the vehicle speed less than 0.3 m / s or the minimum value of the start time.

[0170] The vehicle starts braking: the moment when the brake pedal of the vehicle is pressed for the first time during the cutting-in process.

[0171] The vehicle ends braking: the moment when the vehicle's brake pedal is released for the last time during the cut-in process.

[0172] like Figure 7 As shown, this vehicle is traveling in this lane: the target vehicle starts at the right front or right side.

[0173] This vehicle is driving in this lane: the target vehicle starts at the left front or left side.

[0174] S503 data extraction method defines the key scene to be cut out as follows:

[0175] Definition: The process in which the overlap rate between the target vehicle and the host vehicle is greater than 90% and the lateral speed of the target vehicle is greater than the set value within a continuous period of time, and the final overlap rate is less than 0%.

[0176] Start time: When the target vehicle is traveling in the host vehicle's lane, the lateral speed begins to increase and the final overlap rate is less than 0%, and the lateral speed is greater than 0.3m / s (adjustable).

[0177] End time: The moment when the overlap rate between the target vehicle and the host vehicle begins to be less than 0.

[0178] like Figure 8 As shown, the vehicle is traveling in the same lane: at the end time, the target vehicle is located in front of or to the right of the vehicle.

[0179] The vehicle is driving in the same lane: the target vehicle is located in front of or to the left of the vehicle at the end time.

[0180] S6 data extraction results

[0181] Each scene of the S601 data extraction has a unique identifier, which exists in each row of the extracted data;

[0182] The data extracted in S602 includes all original data information and unifies the time axis.

[0183] S7 data extraction result check

[0184] S701 integrates the extracted samples.

[0185] 1. Select 150 to 180 extracted scene data;

[0186] 2. Number the scene segments that need to be inspected starting from 1 and increasing in sequence;

[0187] 3. According to the sample size and expected accuracy, set the sample size, acceptance limit and rejection limit for the corresponding test round. The expressions are as follows:

[0188] (n1, n2 | Ac1, Re1; Ac2, Re2), where: n1 is the number of samples for the first inspection; n2 is the number of samples for the second inspection; Ac1 is the acceptance limit for the first inspection; Re1 is the rejection limit for the first inspection; Ac2 is the cumulative acceptance limit for the second inspection; Re2 is the cumulative rejection limit for the second inspection.

[0189] S702 Randomly select non-repeating n1 and n2 integers between 1 and N respectively, and their values are the fragment numbers to be inspected;

[0190] S703 Inspect the extracted data fragments according to the numbers, and the inspection process is as Figure 9 shown;

[0191] S705 If the number of non-conformities d1 found in the first inspection sample is greater than or equal to Re1, the inspection result should be considered as rejection;

[0192] S706 If the number of non-conformities d1 found in the first inspection sample is between Ac1 and Re1, the second inspection should be carried out according to the plan, and the number of non-conformities in the two inspections should be accumulated, that is, d1 + d2;

[0193] S707 If the sum of d1 and d2 is less than or equal to the acceptance limit Ac2, the inspection result should be considered as acceptance; if the sum of d1 and d2 is greater than or equal to the rejection limit Re2, the inspection result should be considered as rejection.

[0194] The above has introduced in detail a method for extracting applicable to high-speed autonomous driving function scenarios proposed by the present invention. Specific examples are used in this article to elaborate on the principle and implementation manner of the present invention. The description of the above embodiments is only used to help understand the method and its core idea of the present invention; at the same time, for those of ordinary skill in the art, according to the idea of the present invention, there will be changes in the specific implementation manner and application scope. In summary, the content of this specification should not be construed as a limitation to the present invention.

Claims

1. A method for extracting function scenarios applicable to high-speed autonomous driving, characterized in that, It includes the following steps: Step S1: Fix the data acquisition device on the vehicle and collect relevant data of the vehicle itself and the target vehicle; Step S2: Create a data partition serial number column based on the vehicle speed of the vehicle itself and the number of surrounding target vehicles, and partition the collected data according to the data partition serial number column of the vehicle speed of the vehicle itself and the number of surrounding target vehicles; Step S3: Perform data cleaning on each partition obtained in Step S2; Step S4: Based on the data cleaned in Step S3, process the driving parameters of the vehicle itself to obtain the predicted driving area of the vehicle itself, and mark the objects entering the predicted driving area of the vehicle itself as front targets; Step S5: Classify the high-speed automatic driving function scenarios, and extract data for the following vehicle scenarios, cut-in scenarios, and cut-out scenarios of the high-speed automatic driving function scenarios; Step S6: After integrating the extracted scenario data, randomly select non-repeating n1 and n2 integers between 1 and N respectively. The corresponding values are the test numbers, and the extracted data is tested according to the numbers; 2. The method for extracting a high-speed automatic driving function scenario according to claim 1, wherein In the above-mentioned Step S1, the data acquisition device includes a millimeter-wave radar, an intelligent camera, and a lidar; 3. A method for extracting a high-speed autonomous driving function scenario according to claim 1, characterized in that, In the above-mentioned Step S1, the relevant data of the vehicle itself includes the vehicle speed of the vehicle itself, the lateral acceleration of the vehicle itself, the longitudinal acceleration of the vehicle itself, the steering wheel angle of the vehicle itself, the heading angle of the vehicle itself, the longitude of the vehicle itself, and the latitude of the vehicle itself; The relevant data of the target vehicle includes the relative longitudinal distance between the target vehicle and the vehicle itself, the relative lateral distance between the target vehicle and the vehicle itself, the relative speed between the target vehicle and the vehicle itself, the absolute speed of the target vehicle, the absolute lateral acceleration of the target vehicle, and the absolute longitudinal acceleration of the target vehicle; 4. A method for extracting a high-speed automatic driving function scenario according to claim 1, characterized in that, In the above-mentioned Step S3, performing data cleaning on each partition includes the following steps: Step S301: Perform second-order spline curve interpolation upsampling on the vehicle speed of the vehicle itself, the relative speed between the target vehicle and the vehicle itself, and the absolute speed of the target vehicle data; Step S302: Construct the standard deviation of the Gaussian kernel function, determine the mean value, and then start performing convolution operations; Step S303: Downsample the vehicle speed of the vehicle itself, the relative speed between the target vehicle and the vehicle itself, and the absolute speed of the target vehicle data at equal intervals to their original sampling frequencies; 5. The method for extracting a high-speed automatic driving function scenario according to claim 4, characterized in that, In the above-mentioned Step S303, the convolution operation formula is: Where y(n) is the updated value after filtering, G(i) is the Gaussian kernel function template, and h(n-i) is the original data of the vehicle speed of the host vehicle, the relative speed between the target vehicle and the host vehicle, and the absolute speed of the target vehicle.

6. The method for extracting a high-speed automatic driving function scenario according to claim 1, characterized in that In the above-mentioned Step S4, processing the driving parameters of the vehicle itself to obtain the predicted driving area of the vehicle itself includes the following steps: Step S401: Obtain the driving curvature of the vehicle itself and the steering wheel angle data of the vehicle itself that have been collected; Step S402: Calculate the weighted average of the driving curvature of the vehicle itself and the steering wheel angle data of the vehicle itself. Taking the driving trajectory of the vehicle itself as the origin and extending the forward direction of the vehicle itself as the predicted path of the vehicle itself; Step S403: Translate the predicted path of the vehicle itself 1 / 2 lane width along the radial direction to both sides respectively to obtain the predicted driving area of the vehicle itself; 7. A method for extracting a high-speed autonomous driving function scenario according to claim 1, characterized in that In the above-mentioned Step S5, the following vehicle scenarios include starting following, following at a constant speed, following and accelerating, following and decelerating, following and braking to a stop, overtaking in the same lane, and driving on a curve; 8. A method for extracting a high-speed automatic driving function scenario according to claim 1, characterized in that In the above-mentioned Step S5, the principle of extracting data for the following vehicle scenarios, cut-in scenarios, and cut-out scenarios of the high-speed automatic driving function scenarios is: Taking the relative speed between the target vehicle and the host vehicle as the extraction condition, when the relative speed is in different intervals, the target vehicle and the host vehicle are in different interaction states; Create an identification column for the different interaction states of the target vehicle and the host vehicle. When the relative speed > 0, the identification column value is 0, otherwise it is 1; Derive the identification column for the different interaction states of the target vehicle and the host vehicle to obtain the derivative column of the identification column, and set the last value to 1; Derive the identification column for the different interaction states of the target vehicle and the host vehicle in the reverse direction to obtain the reverse derivative column of the identification column, and set the first value to 1; Filter the time points corresponding to the non-zero values in the derivative column and the reverse derivative column. Then, the time axis corresponding to the non-zero points in the reverse derivative column and the zero points in the derivative column forms the start and end times of each relative speed interval; The relative speed interval of the odd order forms the state of the host vehicle corresponding to the first interval, and the even order is the opposite; When the interval between two consecutive identical states is less than 1 s, they are merged; The initial actions of both the target vehicle and the host vehicle are to follow the line.

9. A method for extracting a high-speed automatic driving function scenario according to claim 1, characterized in that In step S6, the integration of the extracted scenario data includes the following steps: Step S601, select 150 - 180 extracted scenario data; Step S602, number the scenario data to be tested incrementally starting from 1; Step S603, set the number of scenario data, acceptance limit, and rejection limit for the corresponding test rounds according to the selected amount of scenario data and the expected accuracy rate.

10. A method for extracting a high-speed autonomous driving function scenario according to claim 1, characterized in that In step S6, the inspection of the extracted data according to the number is specifically as follows: If the number of unqualified items d1 found in the first inspection of the scenario data is less than or equal to Ac1, the inspection result is considered to be acceptance; If the number of unqualified items d1 found in the first inspection of the scenario data is greater than or equal to Re1, the inspection result is considered to be rejection; If the number of unqualified items d1 found in the first inspection of the scenario data is between Ac1 and Re1, conduct a second inspection according to the plan, and accumulate the number of unqualified items in the two inspections, that is, d1 + d2; If the sum of d1 and d2 is less than or equal to the acceptance limit Ac2, the inspection result is considered to be acceptance; If the sum of d1 and d2 is greater than or equal to the rejection limit Re2, the inspection result is considered to be rejection.

Images