A method for storing validation data to verify a system for processing large amounts of data.
Patent Information
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- ROBERT BOSCH GMBH
- Filing Date
- 2026-01-21
- Publication Date
- 2026-08-03
Smart Images
Figure 2026125605000001_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the verification of systems for processing large amounts of data. Modern systems for processing a significant amount of data, such as a driving assistance system, today process a large number of sensor data and can execute complex functions based on these sensor data. In these functions, (in some cases significant) intervention in the interface between the vehicle driver and the vehicle or vehicle functions is performed. Such a function can be, for example, an emergency braking assistance function that intervenes when an obstacle appears during road traffic.
Background Art
[0002] In these types of systems, the evaluation of sensor data is often performed by very complex software systems. Before such a system can be used in the field (i.e., in normal operation), extensive verification is usually required. This is usually true even when only (relatively minor) adjustments are made to an already established system.
[0003] In the case of driver assistance systems, generally applicable standards require thousands of hours of normal driving operation of a vehicle equipped with a particular driver assistance function to obtain approval for field use of each function. Given the large number of driver assistance functions used today and the complexity of these functions, this is virtually impossible to achieve. Therefore, approaches such as performing verification of driver assistance functions under specific conditions in a simulation environment, and / or performing verification of driver assistance functions by inputting stored verification data into the control unit under test and having it perform each driver assistance function, are becoming widespread today. In such a simulation environment or control unit test device, a large amount of (particularly labeled) sensor data is used to test the driver assistance system and, with respect to that, to check whether the driver assistance system responds correctly in the situations represented by the sensor data. Such sensor data that can be used in a simulation environment to test a driver assistance system is usually called verification data. Such verification data must typically include original sensor data, such as that recorded in the vehicle during test driving. Typically, the requirements for such verification data regarding their originality are still much stricter than those for training data, which is used "only" for training driver assistance functions. Training data is often generated entirely synthetically. Training data is typically generated, for example, by filtering the data to create additional training datasets. The requirements for validation data are stricter than those for training data because validation grants approval for the use of driver assistance features in road traffic. Therefore, validation based on validation data is often performed to validate the training of driver assistance features or control units for driver assistance features based on training data.
[0004] To validate the numerous and diverse driver assistance systems available on the market, an entire data center with sufficient storage space for validation data is required. To reduce the resource costs associated with storing and providing such verification data, it is generally highly desirable to be able to use compression methods on such data. Possible methods are described, for example, in German Patent Application Publication No. 102019214587, U.S. Patent No. 11356579, or European Patent Application Publication No. 3185555.
[0005] However, conventionally, it has not been possible to compress such data used to verify the functionality of driver assistance systems using an efficient approach. To meet the usual standards for verifying driver assistance systems, the data used must typically match bit by bit from the data provided by the vehicle or the sensors of the test vehicle. Therefore, verification data can only be lossily compressed if it is also processed in a lossy compression manner in a real vehicle. In a vehicle, such compression has drawbacks, including additional delays for compression and decompression, the non-use of CRC checksums to ensure end-to-end data integrity, and the cost of mass production associated with compression. [Prior art documents] [Patent Documents]
[0006] [Patent Document 1] German Patent Application Publication No. 102019214587 Specification [Patent Document 2] U.S. Patent No. 11356579 [Patent Document 3] European Patent Application Publication No. 3185555 [Overview of the project] [Problems that the invention aims to solve]
[0007] Given this background, the object of the present invention is to mitigate, or at least partially solve, the problems described in the prior art. In particular, a solution should be presented that enables a higher degree of compression of verification data, thereby significantly reducing the resource costs associated with storing and providing verification data. [Means for solving the problem]
[0008] Here, we describe a method for storing verification data in a data storage system. The verification data is provided for the verification of a system for processing large amounts of data. The system for processing large amounts of data may preferably be the driver assistance system mentioned at the beginning. Alternatively, the system for processing large amounts of data may be a medical system, a manufacturing system, or a system for performing manufacturing tests in which very high-resolution cameras are used.
[0009] The data storage system is further designed to provide validation data to verify the system's ability to process at least one large amount of data, and this validation data includes at least some data recorded by sensors. For example, in the case of a driver assistance system, this could be data recorded by sensors in the vehicle during a test drive.
[0010] This method includes the following steps. a) Steps to receive raw sensor data, b) A step of determining at least one statistical characteristic of the received sensor raw data, c) A step of compressing the raw sensor data in order to generate compressed sensor data using an irreversible compression method, d) A step of determining at least one statistical comparison characteristic of the compression validation data, e) A step of performing a comparison of at least one statistical characteristic with at least one statistical comparison characteristic, wherein the comparison checks whether the deviation between the synthesized sensor data that can be generated based on the compressed sensor data and the raw sensor data received in step a) is within a given tolerance, wherein the tolerance is selected such that deviations within the signal noise caused by the characteristics of at least one sensor that recorded the raw sensor data are captured, and the tolerance is further selected such that the distance between the synthesized sensor data and the captured reality is less than or equal to the distance between the raw sensor data received in step a) and the same reality, and f) If the deviation is calculated to be within a given tolerance range in step f), the step of storing the compressed sensor data in the data storage system (12) as part of the verification data.
[0011] The at least one statistical comparison characteristic in step d) is preferably the same as the at least one statistical characteristic in step b). The lossy compression method performed in step c) is preferably two-part, and more preferably includes the following two substeps:
[0012] c1) A step of reducing noise components in the raw sensor data to generate cleaned raw sensor data, and c2) A step of compressing the raw sensor data using a lossless, reversible compression method.
[0013] In particular, if step e) determines that the deviation of at least one statistical characteristic is within a given tolerance, then step f) is performed to store the compressed sensor data in the data storage system.
[0014] The advantage of the method according to the present invention lies in the possibility of obtaining sensor data, particularly compressed sensor data, as part of the verification data, at a far less resource cost than previously required, and these sensor data have the same deviation and therefore the same quality as the raw sensor data typically captured by at least one sensor. In other words, this provides the advantage and proof that the compressed or restored image is not significantly different from reality compared to another image captured from the same reality by a noisy sensor. Here, the aforementioned comparison, particularly regarding tolerances, contributes to the sensor data being qualitatively suitable for verification by driver assistance systems after being stored in a data storage system. In doing so, the sensor data also takes into account the legal requirements necessary for this purpose. Furthermore, considering driver assistance systems, this method can avoid the drawbacks of compression in vehicles.
[0015] According to one embodiment of this method, statistical comparison characteristics are determined in step d) using a verified physical model of at least one sensor. In particular, this is a verified physical model of at least one sensor at each operating point. In this case, at least one statistical characteristic of the sensor preferably corresponds to at least one statistical characteristic of the received raw sensor data. Preferably, at least one statistical comparison characteristic corresponds to a characteristic also possessed by the synthetic sensor data reconstructed based on the compressed verification data.
[0016] To prove that the statistical comparison characteristics obtained by the validated physical model correspond to the statistical characteristics of the sensor, the validated physical model of at least one sensor is validated against at least one sensor, preferably according to defined criteria.
[0017] It is particularly preferable that the validated physical model for determining statistically comparable characteristics includes operational data from at least one sensor. The operation data of at least one sensor is preferably required to set the respective operating points of the at least one sensor in a verified physical model. Preferably, these operation data correspond to raw sensor data.
[0018] It may be further advantageous if at least one statistical characteristic and at least one statistical comparison characteristic are each a distribution of gray-scale values or luminance values in the received raw sensor data.
[0019] This relates in particular to the image data (in some cases individual images) contained in the raw sensor data. In particular, at least one statistical characteristic and at least one comparison characteristic may be the standard deviation of the gray-scale value or luminance value. Further statistical characteristics that can be considered within the framework of at least one statistical characteristic can be the average gray-scale value / luminance value of the raw sensor data (especially the individual images contained in the raw sensor data). Preferably, at least one statistical characteristic is determined over all pixels of the raw sensor data configured as an image, for example, so that each pixel is represented by at least one statistical characteristic and is jointly evaluated via at least one statistical characteristic.
[0020] It is particularly preferred if the raw sensor data is preferably recorded by at least one image sensor, ultrasonic sensor, lidar sensor, and / or radar sensor, with at least one camera.
[0021] In an example of an ADAS, for example, raw sensor data can be recorded during a test drive. The advantage of this embodiment lies in that the described method is applicable to various types of sensors, and thus to raw sensor data. Furthermore, raw sensor data of various sensor types can be used to collect various types of data for verification, thereby improving the quality of the method as a whole, and thus the quality of the generated verification data including compressed sensor data.
[0022] The reception of the raw sensor data in step a) represents, inter alia, the process by which a device designed to execute the described method receives verification data or raw sensor data from the sensors used. This device can be mounted, for example, on a vehicle used to execute the described method. In such a device, further steps of the described method, especially steps b), c), d), and e) can also be executed. In some cases, the generation of raw sensor data by the sensors can also be a component of the method, and thus, the reception of the raw sensor data in step a) may, inter alia, include the generation of raw sensor data by at least one sensor. These raw sensor data can then be processed directly using the described method. In a variant embodiment, it is also possible for the raw sensor data to be temporarily stored before step a) and then "received" from the buffer in step a).
[0023] According to a further embodiment, the verification data includes, in addition to the sensor data, further data recorded in the system in parallel with the raw sensor data. In an example of a driving assistance system, further data can be recorded in the vehicle during a test drive. Such further data can be, for example, in-vehicle data regarding the state of the driver and / or passengers of the vehicle. For example, the data can be image data from an in-vehicle camera that captures the driver's state of concentration. By using these additional data, the quality of the verification data can be further improved.
[0024] According to one embodiment, it can be further advantageous if the tolerance range used in step e) is selected such that deviations within the shot noise of the raw sensor data received in step a) are captured.
[0025] The term "shot noise" in the present invention may mean a special noise pattern that occurs when photons are converted into electrons in a sensor and causes pixel values to change according to a Poisson distribution, and the magnitude of the change depends on the exposure intensity.
[0026] This embodiment ensures that the compressed sensor data is of sufficient quality to enable the generation of sensor data that will be synthesized from it (later). It is even more advantageous if at least one sensor used to obtain the raw sensor data has the feature of switching between different exposure times for each pixel in response to light incidence, and the tolerance used in step e) and / or the comparison characteristics obtained in step d) are determined taking this feature into consideration.
[0027] This embodiment is based on an approach in which the aforementioned tolerance ranges take into account pixels of different brightness or darkness, and thus can be more accurately adapted to the raw sensor data. By feeding the operating data of at least one sensor into a validated physical model, different exposure times can be taken into account in order to create a validated physical model.
[0028] Furthermore, it is preferable to ensure that, for each pixel of sensor data in the verification data, the composite sensor data obtained from the verification data stored in step g) is as close to reality as the original raw sensor data received in step a), by adhering to the tolerance range used in step e).
[0029] This standard ensures that the sensor data obtained from the verification data stored in step f) can be considered a direct representation or counterpart to the raw sensor data, thereby allowing the sensor data to be used for verification of at least one system. This is possible at any small, different point in time, and thus these data can also be used to check systems where safety is critical and / or AI-based systems, in which higher safety levels and requirements apply than in the case of "pure" comfort systems.
[0030] The ability to check compliance with tolerance limits for each pixel is even more advantageous if it allows use in systems that are susceptible to even slight changes in noise behavior. This ensures the already mentioned applicability of this method to driver assistance systems and / or AI-based driver assistance systems where safety is critical, thereby making the requested and stored sensor data usable in a wider variety of systems as validation data.
[0031] Furthermore, it is preferable that the system is a driver assistance system and that at least steps a) to e) are performed in the vehicle during a test drive to record raw sensor data for creating verification data. However, as a supplement, the aforementioned creation of verification data can utilize not only test drives but also, for example, normal driving by employees using company vehicles. This advantageously increases the amount of usable verification data that can be used to verify at least one driver assistance system.
[0032] To store the verification data long-term, step g) includes transmitting the compressed sensor data as part of the verification data to a fixed data storage system. In the example of a driver assistance system, the transmission may be made from a vehicle to perform a test run. Alternatively, wireless transmission of the data via a wireless interface is also possible. Preferably, the transmission of the compressed sensor data is done as part of the verification data, but in the form of a hard disk containing the compressed sensor data, which is physically removed from the system (e.g., the vehicle on which the method is performed). The hard disk is then supplied to the fixed data storage system. The fixed data storage system may be, for example, a fixed hard disk farm on which the data is stored.
[0033] Furthermore, it is preferable that in step e), a comparison protocol is created to record that a given tolerance range is observed, and in step f), the comparison protocol is stored together with the compressed sensor data.
[0034] This comparison protocol is advantageous in that it helps to prove that the compressed verification data is suitable and approved for the verification of at least one driver assistance system. A method for verifying driver assistance systems, a. A step of receiving verification data, including compressed sensor data, from a data storage system, b. Steps to create synthesized sensor data from compressed sensor data, c. The step of performing a test of the driver assistance system based on verification data and synthetic sensor data, Here, the verification data received in step a) is stored in the data storage system according to the method already described above.
[0035] A system for verifying at least one driver assistance system, designed to perform the methods described above, is also described herein. The advantages and preferred forms described regarding the method are applied to the system in a meaningful way, and vice versa.
[0036] The solutions and their technical environment presented herein will be described in more detail below with reference to the drawings. It should be noted that the present invention should not be limited by the illustrated exemplary embodiments. In particular, unless expressly otherwise specified, it is possible to extract aspects of the facts described in the drawings and combine them with other drawings and / or other components and / or findings from this specification. A schematic example is shown below. [Brief explanation of the drawing]
[0037] [Figure 1] This figure shows a method for storing verification data in an example of a driver assistance system, wherein a direct comparison is made between received data and synthesized data. [Figure 2] This figure shows a first modified embodiment of the method described herein, in which the static characteristics of the raw sensor data are examined. [Figure 3] This figure shows a second modified embodiment of the method described herein, in which the statistical characteristics of the raw sensor data are examined. [Figure 4] This figure shows a method for verifying a system for processing large amounts of data in an example of a driver assistance system using such verification data. [Modes for carrying out the invention]
[0038] In drawings, identical or identical components are always represented by the same reference numeral. Figure 1 shows a block diagram of the method for providing the verification data described herein. Figure 1 will be explained first. Figures 2 and 3 show the method described herein as an advanced form of the method presented according to Figure 1. The descriptions of Figures 2 and 3 will highlight the differences from Figure 1, or the differences between Figure 2 and Figure 3. The description of Figure 1 will also apply to Figures 2 and 3 accordingly (to the extent applicable).
[0039] In this method, the real environment 1 is captured by sensor 2. Here, sensor 2 may be, for example, an image sensor, an ultrasonic sensor, a LiDAR sensor, and / or a radar sensor. Preferably, sensor 2 is a camera that captures image data from the real environment 1 as sensor raw data 3. However, sensor 2 may also be a sensor in a medical system (e.g., a diagnostic system), which captures a large amount of data used, for example, for medical diagnostic purposes. Basically, the method for providing verification data described herein is suitable for a wide variety of systems for processing large amounts of data, and here we will describe the method based on a driver assistance system.
[0040] The real-world environment 1 captured by sensor 2 is, on the one hand, transmitted to the driver assistance system 4 in the form of raw sensor data 3, where it can be used for driver assistance functions 5. On the other hand, the raw sensor data 3 is also compressed 6 and stored in the data storage system 12 of the data center 15 as compressed sensor data 11, which is part of the verification data 23, via the transmission interface 13. The transmission interface 13 may be a wireless interface. Alternatively, the data can be stored in the data storage system 12 on a hard drive, for example, without transmission via the transmission interface 13. For example, the hard drive containing the data can be (physically) removed from the test vehicle, brought to the data center, and (physically) read into the data storage system 12.
[0041] In parallel, additional data 22 can be transmitted to the wireless interface 13, and the data 22 is then stored in the data storage system 12 in the data center 15 in the form of verification data 23. Such additional data 22 may be, for example, metadata relating to sensor data. Such additional data 22 may be, among other things, all kinds of data that can be used for labeling sensor data. Alternatively, as already mentioned above, the additional data 22 can also be brought to the data center 15 in the form of a hard drive.
[0042] Providing this type of verification data offers a superior solution in terms of required storage capacity and resource costs compared to conventional technologies. Figure 1 illustrates a method for compressing the verification data (or the raw sensor data 3 within the verification data), and the method described here relies on this method. To better understand the method described here, we will first explain the method shown in Figure 1.
[0043] Within the framework of compression 6, noise reduction 7 is performed first and applied to the raw sensor data 3. The cleaned sensor data 19 that remains after noise reduction 7 is then subjected to reversible lossless compression 8. In other words, the compression of the raw sensor data 3 is preferably performed in two steps. Noise reduction 7 is the first step, and in the first step, information loss occurs in the raw sensor data 3 as noise is removed. Lossless compression 8 is the second step.
[0044] In addition to compression 8, the cleaned sensor data 19 is sent to comparator 9. Here, part of comparator 9 is a synthesizer 24, which adds synthesized noise 21 to the cleaned sensor data 11. The cleaned sensor data 19 (including the added synthesized noise 21) is then compared to the raw sensor data 3 to create a comparison protocol 10. The comparison protocol 10 reflects how well the noisy cleaned sensor data 19 matches the original captured raw sensor data 3. In other words, the comparison protocol 10 indicates whether a given tolerance was observed. The comparison protocol 10 is then preferably stored in the data storage system 12 of the data center 15.
[0045] This approach is further developed by the method for compressing the verification data (or the raw sensor data 3 in the verification data) shown and discussed here in Figures 2 and 3. In particular, a procedure is presented that significantly reduces the computational load for performing the comparison in comparator 9.
[0046] The purpose of the described method is, in particular, to enable the comparison protocol 10 to demonstrate that the quality of the synthesized sensor data 20, created based on the compressed sensor data 11 or the verification data 23, is as good as when the complete raw sensor data 3, captured according to step a), was used without intermediate compression 6 and noise reduction 7.
[0047] The method described herein is based on the recognition that this proof does not need to be performed by a direct comparison of the raw sensor data 3 and the synthesized sensor data 20, and that it is also possible to compare the statistical characteristics 32 of the raw sensor data 3 and the synthesized sensor data 20 with each other in order to perform this proof. Such statistical characteristics 32 of the raw sensor data 3 are generated by statistical data evaluation 31, and the raw sensor data 3 is compared with statistical comparison characteristics 29 in the comparator 9 to create a comparison protocol 10. The comparison protocol 10 can prove that the quality of the synthesized sensor data 20 / verification data created based on the compressed sensor data 11 / verification data is as good as if the captured complete raw sensor data 3 were used without intermediate compression 6 and noise reduction 7.
[0048] Figures 2 and 3 illustrate two different approaches that can determine at least one statistical comparison characteristic 29, and in particular multiple statistical comparison characteristics 29. In Figures 2 and 3, individual method steps a) to f) are assigned to their respective components, and these components perform the individual method steps.
[0049] According to the approach shown in Figure 2, the cleaned sensor data 19 / verification data is fed to the synthesizer 24, which adds synthesized noise 21. In this way, synthesized sensor data 20 is generated. Then, using statistical data evaluation 31, at least one statistical comparison characteristic 29 is generated that can be used to perform a comparison by the comparator 9.
[0050] In parallel with this, the raw sensor data 3 is also sent to the statistical data evaluation 31. Here, at least one statistical characteristic 32 of the raw sensor data 3 is then determined. Subsequently, the determined at least one statistical characteristic 32 can also be used to perform a comparison by the comparator 9. For this purpose, at least one statistical characteristic 32 and at least one statistical comparison characteristic 29 are supplied to the comparator 9. Preferably, at least one statistical characteristic 32 corresponds to at least one statistical comparison characteristic 29, thereby enabling a meaningful comparison. Here, the comparator 9 checks whether the deviation between the synthesized sensor data 20 that can be generated based on the compressed sensor data / verification data 11 and the received verification data (or the raw sensor data 3 in the verification data) is within a given tolerance range. In this case, the tolerance range is selected so as to capture deviations in the signal noise caused by the characteristics of at least one sensor 2 that recorded the verification data, and the tolerance range is further selected so as to be less than or equal to the distance between the synthesized sensor data 20 and the captured reality and the same reality as the raw sensor data 3 in the verification data received in step a).
[0051] After the comparison, a comparison protocol 10 is created to reflect whether the acceptable ranges were observed and whether they were observed. The comparison protocol 10 is then preferably stored in the data storage system 12 of the data center 15.
[0052] The approach presented in Figure 3 is further simplified for practical implementation. Here, at least one statistical comparison characteristic 29 is created using a validated physical model 30 of sensor 2. Using the validated physical model 30, the type and degree of noise 2 included in the sensor raw data 3 by sensor 2 is determined. In other words, according to the modified form in Figure 3, it is ultimately checked whether the noise removed from the sensor raw data 3 by noise reduction 7 matches the type and degree of noise included in the sensor raw data 3 by sensor 2.
[0053] By obtaining at least one statistical comparison characteristic 29 using a validated physical model 30 of sensor 2, a significant simplification of practical implementation is achieved compared to the approach described in Figure 2. This is because it eliminates the need to perform a statistical data evaluation 31 of the cleaned sensor data 19 / validation data supplied to the synthesizer 24. Similarly, the addition of synthesized noise 21 is also eliminated. Preferably, the validated physical model 30 accepts at least one operating data 33 of sensor 2 as an input variable. A validated physical model 30 using at least one operating data 33 of sensor 2 is particularly preferred so that the noise behavior of sensor 2 can be reproduced in order to create the statistical comparison characteristic 29.
[0054] The motion data 33 may, in particular, be the luminance values of the data (image) captured by the sensor. The motion data 33 may, in particular, correspond to the raw sensor data 3, and then, preferably, the grayscale / luminance values (only) of the raw sensor data 3 are used to determine the statistical comparison characteristics 29 by the verified physical model. Particularly preferably, this is done individually, periodically and repeatedly (particularly preferably for each image and / or a large number of images included in the raw sensor data). Even more preferably, this is done on a pixel-by-pixel basis, individually for each pixel and / or group of pixels in the image in the raw sensor data. Thus, different exposure times of pixels in the image in the raw sensor data can be taken into account to determine at least one statistical comparison characteristic. Figure 4 shows a block diagram of a method for verifying a system for processing large amounts of data in an example of a driver assistance system using verification data obtained according to the method described with respect to Figures 1, 2, and 3. Steps b. and c. of this method are marked in Figure 4. Step a. essentially corresponds to the method for storing the verification data described above with reference to Figures 1 to 3. Step a is shown accordingly in Figures 1 to 3.
[0055] To verify the system for processing large amounts of data, verification data 23 stored in the data storage system 12 is received. In the next step, synthesized sensor data 20 is created 24 from the compressed sensor data 11 in the verification data 23. This is done by adding synthesized statistical characteristics 21, such as synthesized grayscale values, a distribution of synthesized grayscale values, or synthesized noise. The synthesized sensor data 20 is of sufficient quality for verification and matches, for example, the raw sensor data 3 which also always contains noise.
[0056] Next, the sensor data 20, thus "processed" or synthesized, is supplied to the situation simulation or control unit inspection bench 17. The situation simulation or control unit inspection bench 17 is used to simulate or verify the driver assistance system 4. In this case, for example, an emergency braking scenario can be simulated and the driver assistance system 4 can be verified accordingly. In this case, the synthesized sensor data 20 is preferably data, such as image data, that includes the driver assistance system 4 and the scenario in which emergency braking by the driver assistance system 4 should be initiated.
[0057] This simulation or this control unit test bench subsequently produces a verification result 18, which indicates whether the verification of the driver assistance system 4 was successful and provides a value for evaluating the system's performance. [Explanation of Symbols]
[0058] 2 sensors 3. Raw sensor data 6 Compression 7. Noise Reduction / Requirements 11 Compression Sensor Data 12 Data Storage Systems 20. Synthetic Sensor Data 23 Verification Data 29 Statistical comparison properties 30 Verified Physical Models 32 Statistical properties
Claims
1. A method for storing verification data (23) in a data storage system (12), - The aforementioned verification data (23) is provided for the purpose of verifying a system for processing large amounts of data. - The data storage system (12) is designed to provide verification data (23) for verifying a system for processing at least one large amount of data, - The verification data (23) includes at least partially the sensor data (3, 11, 20) recorded by the sensor (2), The method is a) A step of receiving raw sensor data (3), b) A step (7) to determine at least one statistical characteristic (32) of the received sensor raw data (3), c) A step of compressing (6) the raw sensor data (3) in order to generate compressed sensor data (11) using an irreversible compression method, d) A step of finding at least one statistically comparable characteristic (29), e) A step of performing a comparison of the at least one statistical characteristic (32) with the at least one statistical comparison characteristic (29), wherein the comparison checks whether the deviation between the synthesized sensor data (20) that can be generated based on the compressed sensor data and the raw sensor data received in step a) is within a given tolerance range, wherein the tolerance range is selected such that deviations within the signal noise caused by the characteristics of the at least one sensor (2) that recorded the raw sensor data (3) are captured, and the tolerance range is further selected such that the distance between the sensor data (20) synthesized based on the compressed sensor data (11) and the captured reality is less than or equal to the distance between the raw sensor data (3) received in step a) and the same reality, f) If the deviation is calculated in step e) to be within the given tolerance range, the step of storing the compression sensor data (11) in the data storage system (12) as part of the verification data (23), Methods that include...
2. The method according to claim 1, wherein the statistical comparison characteristics (29) are determined in step d) using a verified physical model (30) of at least one sensor (2).
3. The method according to claim 2, wherein the verified physical model (30) is applied to determine the statistical comparison characteristics (29) using the operating data (33) of the at least one sensor (2).
4. The method according to claim 1, wherein the statistical comparison characteristics (29) are the statistical characteristics of the composite sensor data (20) generated from the compressed sensor data (11) generated in step c).
5. The method according to any one of claims 1 to 4, wherein the at least one statistical characteristic and the at least one statistical comparison characteristic are, respectively, the distribution of grayscale values and / or luminance values in the received sensor raw data (3).
6. The method according to any one of claims 1 to 5, wherein the sensor raw data (3) is recorded by at least one image sensor, ultrasonic sensor, lidar sensor and / or radar sensor, preferably at least one camera.
7. The method according to any one of claims 1 to 6, wherein the verification data (23) includes further data recorded in the system in parallel with the sensor raw data (3).
8. The method according to any one of claims 1 to 7, wherein the tolerance used in step e) is selected so as to capture deviations in shot noise of the sensor raw data (3) received in step a).
9. The method according to any one of claims 1 to 8, wherein the at least one sensor (2) for obtaining the sensor raw data (3) is characterized by switching between different exposure times for each pixel in response to light incidence, and the tolerance range used in step e) and / or the comparison characteristic (29) obtained in step d) is determined taking the characteristics into consideration.
10. The method according to any one of claims 1 to 9, wherein, for each pixel of the sensor data (20) in the verification data, by complying with the tolerance range used in step e), the composite verification data (20) obtained from the verification data stored in step f) is made as close to reality as the original raw sensor data (3) received in step a).
11. The method according to any one of claims 1 to 10, wherein the system for processing a large amount of data is a driver assistance system, and at least steps a) to e) are performed in the vehicle during a test run to record sensor raw data (3) for creating verification data.
12. The method according to any one of claims 1 to 11, wherein step f) includes transmitting the compressed verification data (11) to a fixed data storage system (12).
13. The method according to any one of claims 1 to 12, wherein the compression sensor data (11) from step c) is transmitted wirelessly from the system.
14. The method according to any one of claims 1 to 13, wherein in step e) a comparison protocol (10) is created to record that the given tolerance range is observed, and in step f) the comparison protocol is stored together with the compression sensor data (11).
15. A method for verifying a system for processing large amounts of data, a. A step of receiving verification data (23) including compressed sensor data (11) from a data storage system (12), b. A step of creating composite sensor data (20) from the compressed sensor data (11), c. The step of performing a test to verify the system for processing large amounts of data based on the verification data (23) and the synthesized sensor data (20), A method wherein the verification data (23) received in step a) is stored in the data storage system (12) according to the method of any one of claims 1 to 14.
16. A system for verifying a system for processing large amounts of data, designed to perform the method described in claim 15.