Method and system for testing data recording function of DSSAD

Abstract

The embodiment of the invention discloses a DSSAD data recording function test method and system, and relates to the technical field of automatic driving test. The method comprises the steps that real vehicle data and / or simulated driving data of an automatic driving vehicle are / is collected through a data collection tool, and the collected data comprise vehicle state data, driver operation data, automatic driving system data and environment data; a data playback tool is used for verifying and marking the collected data, and the marking content comprises a collision event and a collision risk event; injecting the collected data into the DSSAD to be tested, so that the DSSAD records related data according to standard requirements; and comparing the data recorded by the DSSAD with the marked data, and verifying the data recording function of the DSSAD. Complete testing and verification are provided for the data recording function of the DSSAD, and the testing flexibility is improved.

Description

Technical Field

[0001] The embodiments of the present invention relate to the technical field of autonomous driving testing, and in particular to a method and system for testing the data recording function of a DSSAD. Background Art

[0002] DSSAD (Data Storage System for Automated Driving) has the functions of data loop recording, collision risk event trigger judgment and recording, collision event trigger judgment and recording, and some other functions specified by the standards.

[0003] There is no complete solution for how to test and verify the data logging function of DSSAD. Summary of the invention

[0004] The embodiment of the present invention provides a method and system for testing the data recording function of a DSSAD, which provides complete testing and verification for the data recording function of the DSSAD and improves the flexibility of the test.

[0005] In a first aspect, an embodiment of the present invention provides a method for testing a data recording function of a DSSAD, comprising: Using data collection tools, collect real vehicle data and / or simulated driving data of the autonomous driving vehicle, wherein the collected data includes vehicle status data, driver operation data, autonomous driving system data, and environmental data; Use data playback tools to verify and annotate the collected data, including collision events and events with collision risks; Inject the collected data into the DSSAD to be tested so that the DSSAD records the relevant data according to the standard requirements; The data recorded by the DSSAD is compared with the marked data to verify the data recording function of the DSSAD.

[0006] In a second aspect, an embodiment of the present invention provides a data recording function test system for a DSSAD, comprising: A data collection tool for collecting real vehicle data and / or simulated driving data of an autonomous driving vehicle, wherein the collected data includes vehicle status data, driver operation data, autonomous driving system data, and environmental data; Data playback tool, used to verify and annotate the collected data, where the annotation content includes collision events and events with collision risk; A data injection tool, used to inject the collected data into the DSSAD to be tested, so that the DSSAD records the relevant data according to the standard requirements; The verification and comparison tool is used to compare the data recorded by the DSSAD with the marked data to verify the data recording function of the DSSAD.

[0007] In summary, the embodiment of the present invention provides a method and system for testing the data recording function of DSSAD, and establishes a full test process from test data collection, test data playback, test data injection to comparison and verification of recording results. Each stage is independently completed by a separate tool or device. There is no need to bind DSSAD to a real vehicle or driving simulation system at all times in the traditional way. Instead, the data recording of any vehicle in any period can be tested anytime and anywhere, which improves the flexibility of DSSAD data recording function testing. In addition, through the data collection tool and the simulated driving system, full coverage of collision events, collision risk events and normal data can be achieved; through data playback, preliminary verification and annotation of collected data can be performed to improve the quality of test data; through a variety of comparison and verification mechanisms, the recording of key data by DSSAD can be gradually determined to obtain accurate test results, and provide accurate data entry points (such as inconsistent events) for the improvement of DSSAD, thereby continuously optimizing the data recording function of DSSAD. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0009] Figure 1 It is a structural schematic diagram of a DSSAD data recording function test system provided by an embodiment of the present invention; Figure 2 is a flow chart of a method for testing a data recording function of a DSSAD provided by an embodiment of the present invention; Figure 3 is a schematic diagram of a driving simulation system provided by an embodiment of the present invention; Figure 4 This is a schematic diagram of a collection method of data in different scenarios provided by an embodiment of the present invention; Figure 5 is a flow chart of data collection provided by an embodiment of the present invention; Figure 6 is a flow chart of data playback provided by an embodiment of the present invention; Figure 7 is a schematic diagram of data injection provided by an embodiment of the present invention; Figure 8is a flow chart of data injection provided by an embodiment of the present invention; Fig. 9 This is a flow chart of data comparison and verification provided by an embodiment of the present invention; Fig.10 It is a structural schematic diagram of a driver engagement prediction model provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0010] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work belong to the scope of protection of the present invention.

[0011] In the description of the present invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., indicating the orientation or positional relationship, are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", and "third" are used for descriptive purposes only, and cannot be understood as indicating or implying relative importance.

[0012] In the description of the present invention, it is also necessary to explain that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0013] The embodiment of the present invention provides a method for testing the data recording function of a DSSAD. In order to illustrate the method, a system for testing the data recording function of a DSSAD that supports the implementation of the method is first introduced. Figure 1 FIG. 1 is a structural diagram of a DSSAD data recording function test system provided by an embodiment of the present invention. Figure 1 As shown, the system includes a data acquisition tool, a data playback tool, a data injection tool and a verification and comparison tool.

[0014] Among them, the data acquisition tool is used to collect the real vehicle data or simulated driving data of the autonomous driving vehicle. The real vehicle data comes from the vehicle itself, and the simulated driving data comes from the simulated driving system. The collected data is input into the data playback tool for verification and annotation, and the replayed data can be viewed in the host computer. The annotated data is injected into the DSSAD to be tested through the data injection tool, and the DSSAD records it. The verification and comparison tool is used to compare the data recorded by the DSSAD with the annotated data to verify whether the data recording function of the DSSAD meets the requirements.

[0015] It should be noted that the DSSAD in this embodiment can be a DSSAD integrated inside the autonomous driving vehicle, or it can be a DSSAD tool or box independent of the autonomous driving vehicle, and this embodiment does not make any specific limitations.

[0016] Based on the above system, Figure 2 1 is a flow chart of a method for testing the data recording function of a DSSAD provided by an embodiment of the present invention. The method can be executed by the various parts of the above system in coordination, or by other electronic devices. Figure 2 As shown, the method specifically includes: S110. Collecting real vehicle data and / or simulated driving data of the autonomous driving vehicle through a data collection tool.

[0017] This step collects data, including real vehicle data and simulated driving data. When collecting real vehicle data, the data collection tool can be deployed inside the vehicle to collect environmental data and vehicle data during the actual use of the vehicle, and ensure the synchronization and integrity of the data. Environmental data includes video data collected by the on-board camera, and vehicle data includes vehicle status data (speed, acceleration, speed change, acceleration change, etc.), driver operation data (steering wheel angle, pedal angle, etc.) and autonomous driving system data (autonomous driving system calculation data, decision data, etc.).

[0018] When collecting simulated driving data, you can use third-party software to build a vehicle model and multiple road scenes in the simulated driving system, and display the road scenes to the driver through a display device, so that the driver can drive the car driving simulator under the road scene. The driving operation signal in the car driving simulator will be transmitted to the third-party software to control the movement of the vehicle model. At the same time, DSSAD can store relevant environmental data and vehicle data, such as Figure 3 shown.

[0019] The collected data is divided into three categories: normal driving data, collision risk data, and collision data. The first two types of data can be collected by real vehicles or through simulated driving systems; the third type of collision scenario data can be collected by real vehicles with third-party professional organizations (such as collision experiments), but most of them are collected through simulated driving systems, such as Figure 4 shown.

[0020] In one embodiment, in combination Figure 5 , the specific steps of data collection include: in the data collection tool, the following operations are periodically performed at a frequency higher than the CAN signal and the video signal (that is, the frequency of the data collection tool periodically performing the following steps is higher than the frequency of the CAN signal and the frequency of the video signal): S1-1, determine whether a CAN signal is received; if a CAN signal is received, store the current system time (ie, timestamp), data type, CAN data length, and CAN data, and enter S1-2; if no CAN signal is received, directly enter S1-2.

[0021] S1-2, determine whether the video signal is received; if the video signal is received, store the current system time (ie, timestamp), data type, video data length and video data, and enter the next cycle of the loop; if the video signal is not received, directly enter the next cycle of the loop; know that the data acquisition is completed.

[0022] S120: Using a data playback tool, verify and annotate the collected data, wherein the annotated content includes collision events and events with collision risks.

[0023] Among them, collision events and collision risk events are clearly defined in relevant standards and are also important data recorded by DSSAD. After data collection is completed, this step uses the data playback tool to synchronously playback video data and vehicle data, checks the integrity and relevance of video data and vehicle data before the next step of data injection, and records the triggering time of collision risk events (also called pre-collision events) and collision events to provide true values ​​for subsequent comparison and verification.

[0024] In one embodiment, in combination Figure 6 The specific steps of data playback include: In the data playback tool, at a frequency higher than the CAN signal and the video signal (the same as the frequency in the above data acquisition tool), periodically perform the following operations: S2-1, determine whether the current cycle meets the playback cycle of the CAN signal; If the current cycle meets the playback cycle of the CAN signal (i.e., the time interval meets the CAN playback), send the CAN data of the current frame to the playback interface, and read the timestamp and data of the next frame of the CAN signal; continue to determine whether the current CAN data meets the trigger conditions for recording a collision event or a collision risk event; if the trigger conditions are met, mark the collision event or the collision risk event and enter S2-2; if the trigger conditions are not met, enter S2-2; If the current cycle does not meet the playback cycle of the CAN signal, directly enter S2-2.

[0025] S2-2, determining whether the current cycle meets the playback cycle of the video signal; If the current cycle meets the playback cycle of the video signal (i.e. the time interval meets the video playback), the video data of the current frame is sent to the playback interface, and the current video image is compared with the CAN signal to verify whether the scene is consistent; if they are consistent, the timestamp and data of the next frame of the video signal are read and the next cycle is entered; if they are inconsistent, the inconsistent event is recorded for manual analysis; If the current cycle does not meet the playback cycle of the video signal, it directly enters the loop of the next cycle.

[0026] S130, injecting the collected data into the DSSAD to be tested, so that the DSSAD records the relevant data according to the standard requirements.

[0027] This step performs data injection. The data recording program is started in advance in the DSSAD to be tested, and then the verified and annotated data is injected into the DSSAD through the data injection tool to test the data recording program. Figure 7 shown.

[0028] In one embodiment, in combination Figure 8 The steps of data injection include: the data injection tool determines the data parsing protocol corresponding to the vehicle model according to the vehicle model corresponding to the collected data; according to the data parsing protocol, the collected data is parsed into CAN data and video data; according to the CAN frequency, the CAN data is sent to DSSAD via CAN; according to the video data frequency, the video data is sent to DSSAD via Ethernet; the data recording program running inside DSSAD will record the necessary data.

[0029] S140, comparing the data recorded by the DSSAD with the marked data to verify the data recording function of the DSSAD.

[0030] This step performs data comparison and verification, comparing the data recorded by DSSAD with the true value data marked in S120 to verify whether the data recording function of DSSAD meets the requirements.

[0031] In one embodiment, in combination Fig. 9 ,The specific steps of data comparison and verification include: S4-1. Classify data. Classify the data recorded in the DSSAD into trigger events and continuous records, wherein the trigger events include collision events and collision risk events.

[0032] S4-2, compare the number of trigger events in the recorded data and the marked data to see if they are consistent; if they are consistent, proceed to S4-3; if they are inconsistent, record the inconsistent event names and proceed to S4-3.

[0033] S4-3, comparing in detail whether the recorded data is consistent with the data inside each trigger event in the marked data; if they are consistent, proceed to S4-4; if they are inconsistent, record the inconsistent event name and proceed to S4-4.

[0034] S4-4, comparing the recorded data with the number, duration, and file size of continuous records (such as data during normal driving) in the annotated data to see if they are consistent; if they are consistent, proceed to S4-5; if they are inconsistent, record the inconsistent event name and proceed to S4-5.

[0035] S4-5. Randomly extract a continuous record file (for example, randomly extract a video file) from the recorded data and the marked data, and compare whether the data inside the file is consistent. If they are consistent, randomly extract the next record file until the verification is completed; if they are inconsistent, record the inconsistent event name, and randomly extract the next record file until the verification is completed.

[0036] If the data in the above comparison and verification links are consistent, it means that the data recording function of DSSAD meets the requirements. If there is any inconsistent event, manual analysis of the event is required.

[0037] Further, such as Figure 4 As shown, during the data collection phase, the data of the collision event comes from a third-party professional organization or a simulated driving system. In practical applications, it is more common to use a simulated driving system to generate test cases for collision events. In order to make the simulated collision event closer to the real collision event, this embodiment provides a method to simulate the collision event through the data of a real vehicle with a collision risk event, so as to improve the quality of the simulation data and the test effect of the entire method. In a specific implementation, the process may include the following steps: Step 1: Use data collection tools to collect collision risk events from real vehicle data.

[0038] Step 2: Generate vehicle status data and environmental data in a simulated driving system based on specific real vehicle data within a set time period before the collision risk event, so that the driver can control the autonomous driving vehicle model through a car driving simulator to simulate collision events under the specific real vehicle data multiple times.

[0039] For example, the relevant standards require DSSAD to record data 15 seconds before the start of a collision risk event. Here, the set time length can be set to be greater than 15 seconds, such as 30 seconds, and the real vehicle data 30 seconds before the start of the collision risk event can be recorded in advance. For the convenience of distinction and description, the real vehicle data of 30 seconds is called specific real vehicle data.

[0040] Then, the environmental data in the specific real vehicle data is used as the road scene of the vehicle driving in the simulated driving system, the vehicle state data for a period of time in the specific real vehicle data is used as the initial vehicle state in the simulated driving system, and some parameters of the automatic driving system data for a period of time in the specific real vehicle data (such as system setting parameters with less real-time variability, etc.) are used as some initial parameters of the automatic driving system in the simulated driving system to generate a virtual driving scene. The driver can simulate driving through the car driving simulator, control the movement of the vehicle model, and simulate multiple collision events in the scene. Since the environmental data, vehicle state data, and some parameters of the automatic driving system of these collision events are generated based on the collision risk events in the real vehicle data, the similarity between the simulated scene and the real vehicle collision scene can be basically guaranteed, so that the overall simulated collision event is as close to the real collision event as possible.

[0041] Step three, calculate the driver engagement in the specific real vehicle data and the driver engagement in the early stage of each simulation data; compare the driver engagement in the early stage of each simulation data with the driver engagement in the specific real vehicle data, and select a simulation data with the most similar driver engagement as the final simulated collision event.

[0042] As described above, in the multiple collision events simulated in step 2, the vehicle state data, environmental data and part of the data of the autonomous driving vehicle are very close to the real scene. The only thing with strong uncertainty is the driver's response. Different drivers or the same driver have driving operation deviations in different simulations, and the driver's intentional collision in the simulated collision event and the driver's avoidance of collision as much as possible in the real scene will also lead to driving operation deviations. In order to minimize this deviation, this embodiment calculates the driver's participation in the specific real vehicle data (representing the real participation of the person) and the driver's participation in the simulated data in the same period, compares the two, and selects the simulation data that is most similar to the driver's participation in the real vehicle data as the simulated collision event that is closest to the actual situation.

[0043] Specifically, the driver engagement in this embodiment characterizes the impact of the driver's operation on the vehicle's driving, which is related to both the driver's subjective desire to exert control over the vehicle and the interaction between the driving operation and the automatic driving system. Therefore, this embodiment also considers the impact of the automatic driving system on the vehicle's driving in the calculation of the driver engagement.

[0044] In a specific implementation, a driver engagement prediction model based on a neural network may be pre-built, such as Fig.10 As shown, the model includes two feature converters, one of which is used to convert the driver operation data vector into an embedding vector with the same dimension as the vehicle state data vector, and the other is used to convert the automatic driving system data vector into an embedding vector with the same dimension as the vehicle state data vector, and the two embedding vectors are linearly independent. Optionally, each feature converter can adopt a fully connected layer or a convolutional network, etc.; the two feature converters are independent of each other, and the structure and parameters are not necessarily related. Among them, the driver operation data vector can be generated by the following method: dividing the specific real vehicle data into M time slices (M is a natural number), arranging the N types of driver operation data of the same time slice into a column vector, and splicing the column vectors of all time slices into a driver operation vector (M´N dimension); or, performing principal component analysis on the column vectors of all time slices to obtain at least one N-dimensional principal component vector, and then performing weighted summation on these principal component vectors according to the corresponding eigenvalues ​​to obtain an N-dimensional driver operation data vector. The automatic driving system data vector is similar to the vehicle state data vector, as long as the generation method of the three data vectors is consistent. Usually, the dimensions of the three data vectors are different, and their respective dimensions are related to the number of parameters collected.

[0045] Based on the above model structure, this embodiment uses the vehicle state data, driver operation data and automatic driving system data belonging to the same time period (the time period length is the set time length) as a sample, and annotates the driver participation and intelligent driving system participation in each sample, wherein the intelligent driving system participation is used to characterize the impact of the intelligent driving system on vehicle driving. Optionally, the driver participation can be jointly annotated by the driver and the automatic driving system expert, taking into account the driver's subjective expectation of control over the vehicle and the objective impact on the vehicle state, and the intelligent driving system participation is annotated by the automatic driving expert, and both participations are values ​​with an absolute value less than 1.

[0046] After the samples are labeled, the model is trained using the sample set so that the driver operation data vector and the autonomous driving system data vector belonging to the same sample are respectively input into two feature converters, and then the two embedded vectors are weightedly summed using the labeled driver engagement and autonomous driving system engagement to obtain the vehicle state data vector of the same sample.

[0047] Specifically, the following loss function can be constructed L :

[0048] in, represents the embedding vector obtained after the driver operation data vector passes through one of the feature converters, represents the embedding vector obtained after the autonomous driving system data vector passes through another feature converter, Represents the vehicle state data vector; , and All are from the same sample; Indicates that and After being expressed as a column vector, the rank of the matrix formed by arranging two column vectors is, Indicates absolute value; They represent the annotated driver engagement and autonomous driving system engagement, respectively; are the preset weight coefficients respectively.

[0049] During the training process, L To update the model parameters, we minimize L At minimum, Approaching 0, , used to ensure and is nonlinearly related. , and have the same dimensions, so this embodiment will Expressed as , The linear weighting of to reflect the joint effect of the driver and the autonomous driving system on the vehicle state; L At minimum, Approaching 0, , which happens to reflect the driver’s engagement and the autonomous driving system’s engagement in the vehicle status in a linearly related way.

[0050] Furthermore, the following loss function can be constructed L 1:

[0051] in, , and The meaning of is the same as that of formula (1), and both are from the same sample; They respectively represent the engagement of the autonomous driving system marked in other samples different from the same sample and the embedding vector obtained after the autonomous driving system data vector passes through the feature converter, Represents the index of other samples, represents the sum of all other samples different from the same sample; are the preset weight coefficients respectively. During the training process, To update the model parameters, we minimize L 1 minimum, Very large, the three data vectors used to constrain different samples cannot satisfy the above linear relationship, and the loss term The addition of can speed up the convergence of the model and improve the training effect.

[0052] After the training is completed, in the model use phase, the driver operation data vector in the specific real vehicle data is and the autonomous driving system data vector , respectively input two trained feature converters, and obtain two new embedding vectors Then, the vehicle data vector in the specific real vehicle data is It is expressed as the weighted sum of the two new embedding vectors , where the new embedding vector corresponding to the driver's operation data vector is Weight That is, the driver engagement in the specific real vehicle data.

[0053] Similarly, for any collision event simulated in step 2, the driver operation data, automatic driving system data and vehicle status data in the same period as the specific real vehicle data are extracted from the full simulation data to generate the corresponding driver operation data vectors , autonomous driving system data feature vector and the vehicle status data vector .Will and Input two trained feature converters respectively to obtain two new embedding vectors Then, Expressed as The weighted sum form of ,in, Weight This is the driver’s involvement in this simulated collision event.

[0054] The same operation is performed on all simulated collision events, and the driver engagement of each simulated collision event can be obtained. A simulated collision event that is closest to the driver engagement in the specific real vehicle data is selected as the final simulated collision event. This collision event is generated based on the collision risk event in the real vehicle data, and the driving scene in the actual event (including environmental data, initial vehicle status and some parameters of the autonomous driving system, etc.) is used, and the driver engagement that is most similar to the actual event is maintained, and the actual situation is restored to the greatest extent in all aspects, so that the subsequent simulated collision events are closer to the real collision events, which improves the objective authenticity of the test data, thereby improving the accuracy and objectivity of the entire test method.

[0055] The above-mentioned method of generating simulated collision events can be executed by a processing chip deployed inside the data acquisition tool, or by an electronic device outside the data acquisition tool, or by a data acquisition tool and an independent electronic device in cooperation with each other, which is not specifically limited in this embodiment.

[0056] In summary, this embodiment provides a method for testing the data recording function of DSSAD, and establishes a full test process from test data collection, test data playback, test data injection to comparison and verification of recording results. Each stage is completed independently by a separate tool or device. There is no need to bind DSSAD to a real vehicle or driving simulation system at all times in the traditional way. Instead, the data recording of any vehicle in any period can be tested anytime and anywhere, which improves the flexibility of DSSAD data recording function testing. In addition, through data collection tools and simulated driving systems, full coverage of collision events, collision risk events and normal data can be achieved; through data playback, preliminary verification and annotation of collected data can be performed to improve the quality of test data; through a variety of comparison and verification mechanisms, the recording of key data by DSSAD can be gradually determined to obtain accurate test results, and provide accurate data entry points (such as inconsistent events) for the improvement of DSSAD, thereby continuously optimizing the data recording function of DSSAD.

[0057] like Figure 1 As shown, the embodiment of the present invention also provides a DSSAD data recording function test system. The data collection tool is used to collect real vehicle data and / or simulated driving data of the autonomous driving vehicle, wherein the collected data includes vehicle status data, driver operation data, autonomous driving system data and environmental data.

[0058] The data playback tool is used to verify and annotate the collected data, where the annotated content includes collision events and events with collision risk.

[0059] The data injection tool is used to inject the collected data into the DSSAD to be tested so that the DSSAD records the relevant data according to the standard requirements.

[0060] The verification and comparison tool is used to compare the data recorded by the DSSAD with the annotated data to verify the data recording function of the DSSAD.

[0061] It should be noted that this embodiment and the above method embodiment are based on the same inventive concept, and the limitations in any of the above method embodiments are applicable to this embodiment and can achieve the same beneficial effects.

[0062] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the technical solutions of the embodiments of the present invention.

Claims

1. A method for testing the data recording function of a DSSAD, characterized in that: include: Using data collection tools, collect real vehicle data and / or simulated driving data of the autonomous driving vehicle, wherein the collected data includes vehicle status data, driver operation data, autonomous driving system data, and environmental data; Use data playback tools to verify and annotate the collected data, including collision events and events with collision risks; Inject the collected data into the DSSAD to be tested so that the DSSAD records the relevant data according to the standard requirements; The data recorded by the DSSAD is compared with the marked data to verify the data recording function of the DSSAD.

2. The method according to claim 1, characterized in that: The collecting of the real vehicle data and / or simulated driving data of the autonomous driving vehicle by the data collection tool includes: In the data acquisition tool, the following operations are performed periodically at a frequency higher than that of the CAN signal and the video signal: S1-1, in response to receiving a CAN signal, storing a timestamp, a data type, a CAN data length, and CAN data; S1-2. In response to receiving a video signal, store a timestamp, a data type, a video data length, and video data.

3. The method according to claim 1, characterized in that The data playback tool is used to verify and annotate the collected data, including: In the data playback tool, perform the following operations periodically at a frequency higher than that of the CAN signal and the video signal: S2-1, in response to the current cycle satisfying the playback cycle of the CAN signal, sending the current CAN data, and reading the timestamp and data of the next frame of the CAN signal; if the CAN data satisfies the triggering conditions of the collision event or the collision risk event, marking the collision event or the collision risk event; S2-2, in response to the current cycle satisfying the playback cycle of the video signal, sending the current video data, and reading the timestamp and data of the next frame of the video signal; comparing the current video image with the CAN signal to verify whether the driving scene is consistent.

4. The method according to claim 1, characterized in that: The collected data is injected into the DSSAD to be tested so that the DSSAD records relevant data according to the standard requirements, including: Determine the data analysis protocol based on the vehicle model of the collected data source; According to the data analysis protocol, the collected data is analyzed into CAN signals and environmental data; According to the frequency of the CAN signal and the environmental data, the CAN signal and the environmental data are sent to the DSSAD to be tested so that the DSSAD records the relevant data according to the standard requirements.

5. The method according to claim 1, characterized in that The step of comparing the data recorded by the DSSAD with the marked data to verify the data recording function of the DSSAD includes: The data recorded in the DSSAD is divided into trigger events and continuous records, wherein the trigger events include collision events and collision risk events; Compare the recorded data with the number of trigger events in the marked data to see if they are consistent, and record the names of inconsistent events; Compare the recorded data with the data in each trigger event in the marked data to see if they are consistent, and record the names of inconsistent events; Compare the recorded data with the number, duration, and file size of the continuous records in the marked data to see if they are consistent, and record the names of the inconsistent events; Randomly extracting continuous record files from the recorded data and the marked data, comparing whether the data in the files are consistent, and recording the names of inconsistent events; Among them, the inconsistent event names were used for the final manual analysis.

6. The method according to claim 1, characterized in that The collecting of the real vehicle data and / or simulated driving data of the autonomous driving vehicle by the data collection tool includes: Use data collection tools deployed on real vehicles to collect collision risk events in real vehicle data; Generate vehicle status data and environmental data in a simulated driving system based on specific real vehicle data within a set time period before the collision risk event, so that the driver can control the autonomous driving vehicle model through a car driving simulator to simulate collision events under the specific real vehicle data multiple times; Comparing the driver engagement level in the early stage of each simulation data with the driver engagement level in the specific real vehicle data; The simulation data with the most similar driver involvement is selected as the final simulated collision event.

7. The method according to claim 6, characterized in that The comparing the driver engagement level in the early stage of each simulation data with the driver engagement level in the specific real vehicle data includes: Constructing a driver engagement prediction model, the model includes two feature converters, the two feature converters are used to convert the driver operation data vector and the autonomous driving system data vector into two embedding vectors with the same dimension as the vehicle state data vector, and the two embedding vectors are linearly uncorrelated; A sample is constructed using vehicle state data, driver operation data, and autonomous driving system data belonging to the same time period, and the model is trained so that after the driver operation data vector and the autonomous driving system data vector belonging to the same sample are respectively input into two feature converters, the two embedding vectors are weighted summed using the labeled driver engagement, so that the vehicle state data vector of the same sample can be obtained; The driver operation data vector and the automatic driving system data vector in the specific real vehicle data are respectively input into two trained feature converters to obtain two new embedding vectors respectively; the vehicle state data vector in the specific real vehicle data is represented as the weighted sum of the two new embedding vectors, and the weight of the new embedding vector corresponding to the driver operation data vector is used as the driver engagement.

8. The method according to claim 7, characterized in that The method constructs a sample using vehicle state data, driver operation data, and automatic driving system data belonging to the same time period, and trains the model so that the driver operation data vector and the automatic driving system data vector belonging to the same sample are respectively input into two feature converters, and then the two embedding vectors are weighted summed using the labeled driver participation, so as to obtain the vehicle state data vector of the same sample, including: Construct the following loss function L : ； in, They represent the embedding vectors of the driver operation data vector and the autonomous driving system data vector after the feature converter, represents the vehicle state data vector, Represented by a column vector The rank of the matrix formed by the arrangement, Indicates absolute value; They represent the annotated driver engagement and autonomous driving system engagement, respectively; are preset weight coefficients respectively; pass L To update the model parameters, we minimize .

9. The method according to claim 7, characterized in that: The method constructs a sample using vehicle state data, driver operation data, and automatic driving system data belonging to the same time period, and trains the model so that the driver operation data vector and the automatic driving system data vector belonging to the same sample are respectively input into two feature converters, and then the two embedding vectors are weighted summed using the labeled driver participation, so as to obtain the vehicle state data vector of the same sample, including: Construct the following loss function L : ； in, They represent the embedding vectors obtained after the driver operation data vector and the autonomous driving system data vector from the same sample pass through the feature converter, represents the vehicle state data vector of the same sample, Represented by a column vector The rank of the matrix formed by the arrangement, Indicates absolute value; They represent the annotated driver engagement and autonomous driving system engagement, respectively; They respectively represent the degree of engagement of the autonomous driving system marked in other samples different from the same sample, and the embedding vector obtained after the autonomous driving system data vector passes through the feature converter; Indicates the index of other samples; are preset weight coefficients respectively; pass L To update the model parameters, we minimize .

10. The data recording function test system of DSSAD is characterized by: include: A data collection tool for collecting real vehicle data and / or simulated driving data of an autonomous driving vehicle, wherein the collected data includes vehicle status data, driver operation data, autonomous driving system data, and environmental data; Data playback tool, used to verify and annotate the collected data, where the annotation content includes collision events and events with collision risk; A data injection tool, used to inject the collected data into the DSSAD to be tested, so that the DSSAD records the relevant data according to the standard requirements; The verification and comparison tool is used to compare the data recorded by the DSSAD with the marked data to verify the data recording function of the DSSAD.

Images