Space-saving storage of measurement data and processing products

Abstract

Method (100) for storing measurement data (1) and processing products (2) generated therefrom, wherein a graph (3) is created (110) in which • Nodes (4) correspond to the measurement data (1) or parts (1a) of the measurement data (1), or to processing products (2), • each node (4) corresponding to a specific processing product (2) is connected via edges (5) to those parts (1a) of the measurement data (1) or to those processing products (2) from which this specific processing product (2) arises, and • in association with each node (4) corresponding to a specific processing product (2) ◯ this specific processing product (2), and / or ◯ Instructions (2a) for generating this specific processing product (2) from the measurement data (1, 1a) or processing products (2) from which it arises are stored (120) in a memory (6).

Description

[0001] The present invention relates to the storage of measurement data and processing products generated therefrom with the aim of using only as much storage space as is necessary for processing specific questions based on the measurement data and processing products. State of the art

[0002] Vehicle test drives are conducted for the development and testing of driving dynamics systems, driver assistance systems, or systems for at least partially automated driving. During these test drives, measurement data such as camera images, radar data, lidar data, or other sensor data is collected. Depending on the level of detail, a typical test drive can collect a total of approximately 12 GB of measurement data per second. If a fleet of test vehicles is operated and numerous test drives are conducted for different projects, total data volumes in the order of petabytes can accumulate. Providing such amounts of storage space is a significant cost factor.

[0003] The storage requirements are further increased by the fact that subsequent work with the measurement data results in numerous edited versions of the measurement data, some of which build on one another. This is somewhat comparable to the creative editing of photographs, where the original is joined by various other versions to which various filters have been applied. Disclosure of the invention

[0004] The invention provides a method for storing measurement data and the processing products generated therefrom. The measurement data can, for example, be a temporal sequence of • Measured values ​​of at least one measured quantity, and / or • Point clouds and / or video frames whose point or pixel values ​​represent measured values ​​of at least one measurand. For example, video cameras measure light intensities on pixels arranged in a regular grid. Radar and lidar sensors assign one or more measured variables related to this radar or lidar reflection to any point in space from which a reflected radar or lidar beam appears to come, such as an intensity, directional information in the form of one or more angles, or a relative speed between the radar or lidar sensor and the reflecting object. In addition, a variety of other measured data can be recorded, such as steering angle, acceleration forces, or the positions of the brake and accelerator pedals.

[0005] As part of the process, a graph is created. In this graph, nodes correspond to the measurement data or parts of the measurement data, or to the processing products of the measurement data or parts of the measurement data. Each node corresponding to a specific processing product is connected via edges to those parts of the measurement data or to those processing products from which this specific processing product arises.

[0006] In association with each node corresponding to a specific processing product, this specific processing product is stored in a memory. Alternatively, or in combination with this, instructions for generating this specific processing product from the measurement data or processing products from which it originates can be stored in the memory. In this case, storing "in association" with the node can, for example, include storing a pointer or other reference in the graph to the location where the processing product or the instructions are located in the memory.

[0007] It was recognized that this approach makes it easier to determine how long specific processing products, or specific portions of the measurement data, are still needed for further work. The graph clearly shows which additional processing products depend on the specific processing product or specific portion of the measurement data. These additional processing products can no longer be recalculated once the data on which they depend has been deleted. Conversely, this means that a portion of the measurement data, or a processing product, is no longer needed and can be deleted as soon as no more dependent processing products need to be recalculated and there is no longer a retention requirement.For example, there are rules of good scientific practice that require raw data to be retained for a specific period of time to ensure traceability of the results derived from it. Instead of being permanently deleted, data can also be moved to an archive, for example, where access to the data, which is only expected in rare cases, takes significantly longer (e.g., manually retrieving a magnetic tape from a box containing many magnetic tapes in a warehouse). It is therefore particularly possible that very voluminous measurement data have become highly condensed processed products, and only these are required for further work. Intermediate products that were created on the way to these condensed processed products, as well as possibly the original measurement data, can then be deleted.This is somewhat analogous to the fact that when editing videos, the source material is usually many times larger than the final video. As soon as this video no longer needs to be changed, the source material is dispensable. The graph structure proposed here makes it possible to oversee even more complex dependencies, preventing data from being accidentally deleted too early.

[0008] It was further recognized that processing products that are rarely needed do not necessarily need to be stored at full size.

[0009] Instead, it is sufficient to store deterministic instructions for generating the respective processing product from specific measurement data or previous processing products. When the processing product in question is then requested, it can be freshly generated at any time. The downstream processing then sees exactly the same as if the processing product had been stored in its full size. The only trade-off is that executing the deterministic instructions may take longer than retrieving the processing product from a fast memory. However, considering that • certain instructions for generating processing products, such as simply providing a temporal section of the measurement data, can be processed very quickly and • If rarely required processing products can be stored in slower, but more cost-effective storage, this additional effort can be significantly reduced or even reversed.

[0010] In a particularly advantageous embodiment, a data type of the respective portion of the measurement data or the respective processing product is stored in association with portions of the measurement data or with processing products. The data type can, in particular, further specify the abstract data structure as such. For example, a video frame can be specified such that the image is in black and white. The graph can thus contain a general teaching of how measurement data and / or processing results can be converted from one data type to another.

[0011] In a further particularly advantageous embodiment, a library is additionally provided in which, in association with a source data type and • a target data type and / or • a relationship between the source data type and the target data type containing instructions for converting data of the source data type into data of a target data type. Such a library can contain knowledge about possible conversion paths in an even more general form and also contain many more alternatives than a graph created based on specific conversions. For example, the source data type (e.g., "JPG image," "MPG video clip," or "PNG image") and the relationship between the source data type and the target data type (e.g., "black and white") can be stored in the library as keys, which then lead to instructions on how JPG images, MPG video clips, or PNG images can be converted to black and white.

[0012] In another particularly advantageous embodiment, based on the graph and / or using the library, at least one path is determined by which, starting from at least a predetermined portion of the measurement data and / or from at least one predetermined concrete processing product, a processing product of a predetermined target data type can be obtained. Organizing measurement data and processing products with a graph thus not only saves storage space but also makes it possible to meet new, unforeseen requirements for converting the data into specific data types.

[0013] A particularly advantageous strategy for finding at least one conversion path consists in determining a list of properties of the measurement data or the specific processing product that need to be changed on the way to the target data type, based on the specified portion of the measurement data and / or at least one specified concrete processing product, on the one hand, and the specified target data type, on the other. A path with multiple steps can then be determined, with each step changing one property contained in the list. Each step can then be optimized independently without potentially degrading other steps.

[0014] In another particularly advantageous embodiment, several candidate paths are determined in the manner described above. Each of these candidate paths is assigned a score with respect to a given optimization goal. A candidate path with the best score is selected. In this way, the processing of the measured data can be guided in a direction that is advantageous from the perspective of the specific application, provided that a processing result of the same data type is still obtained. Since a specific action can only be performed or not performed, the optimization is a so-called mixed integer problem.

[0015] As explained above, at least one processed product comprises a temporal excerpt from the measured data and / or from a previous processed product. The test drives mentioned at the beginning in particular provide a large proportion of measured data that was routinely expected in this form and does not provide any new insights. However, certain temporal excerpts that relate to specific events are of interest for further analysis. For example, for the test of an electronic stability program (ESP), those time periods are of particular interest in which the vehicle's driving dynamics are at the limit and the ESP must be activated. One and the same measured data can therefore be used for different tests to create a large number of excerpts, all of which were previously saved in full size. Instead, saving only a reference to the source material and time information for the section to be extracted saves a lot of storage space.At the same time, it does not take very long to reconstruct the temporal section ad hoc from the source material if necessary.

[0016] Particularly advantageous is when the time frame covers a specified period before and after the time of an event identified in the measurement data. By filtering for such events, the measurement data from a long test drive can be condensed into a few moments that are important for further analysis.

[0017] Storage space savings can be dynamically managed based on any lifecycle conditions. This means, for example, that it is possible to decide which portions of the measurement data and / or which processing products are still important based on the specific use of measurement data and / or processing products.

[0018] In a particularly advantageous embodiment, in response to the fact that • a time elapsed since the creation of a specific processed product, and / or • If an observed access behavior to this specific processing product fulfills a specified lifecycle condition, this specific processing product is replaced with instructions for generating this specific processing product from the measurement data or processing products from which it originates. In this way, the memory required by infrequently used processing products is freed up and can be used by more frequently used processing products. This does not limit the options for downstream processing. Should the processing product removed from memory be needed again later, it can be regenerated using the deterministic instructions stored in its place.

[0019] In a further advantageous embodiment, in response to the set of processing products generated from a specific portion of the measurement data or from a specific processing product fulfilling a predefined life cycle condition, this specific portion of the measurement data or this specific processing product is deleted. For example, if all processing products that are expected to be required in the context of the respective application have been generated from the portion of the measurement data, the portion of the measurement data is no longer required and can be deleted. The graph structure proposed here makes it easier to check whether all tasks to be performed with a specific portion of the measurement data or with a specific processing product have actually been completed and whether the portion of the measurement data or the processing product can thus be deleted. Once deleted, portions of the measurement data cannot be reconstructed.Once deleted, processing products can only be reconstructed if not only the instructions for their creation are available, but also the source materials required to execute these instructions (i.e., portions of the measurement data or previous processing products). Conversely, for example, parts of the original data can be saved along with instructions on how to reassemble the original data from these parts. For example, these parts can be easier to handle when stored in a file system or transmitted over networks than a single, very large file.

[0020] Lifecycle conditions regarding how frequently certain processing products are required and how long certain portions of the measurement data, or certain processing products, are required at all, can be determined, for example, from a predefined list of queries that must be answered based on the measurement data and the processing products. For many applications, these lists are created in advance as specifications.

[0021] In another particularly advantageous embodiment, storage locations for measurement data and / or specific processing products are determined based on lifecycle conditions and / or data protection constraints for cross-border data exchange. For example, many cloud providers offer storage space in several storage classes, stored on different types of storage media. These storage classes differ in access speeds and, at the same time, in the price per gigabyte. For example, a solid-state disk (SSD) is the fastest, but also the most expensive, non-volatile storage. At the other end of the scale is storage on magnetic tapes in a "tape library." Here, the storage space is cheapest at the cost of having to wait until a tape drive is available for access.The measurement data can also be stored in compressed form and then decompressed ad hoc when it needs to be accessed again.

[0022] The impact of data protection regulations depends on the type of measurement data recorded and also on the manner in which it is processed. Depending on what is done with the measurement data during processing, the processed products may contain less or more personal information worthy of protection than the original measurement data. For example, aggregating, compressing, or simplifying measurement data can result in the processed product allowing fewer conclusions to be drawn about individual people or their behavior. For example, a vehicle's environmental monitoring system can discard all information that is not relevant to traffic and, as a result, only record the legs of pedestrians on sidewalks to determine whether they are crossing into the roadway. Conversely, merging measurement data from multiple data sources can, for example, make it possible to de-anonymize individual people.

[0023] In a further particularly advantageous embodiment, further nodes and edges of the graph represent ordering features for finding or accessing portions of measurement data or specific processing products. In this way, an abstraction layer is introduced via which a specific portion of the measurement data or a specific processing product can be accessed in the same way every time. In particular, for example, an entity accessing a specific processing product does not need to know whether this specific processing product may have to be obtained from an archive storage or generated anew using deterministic instructions. It should be noted, however, that access may take somewhat longer in these cases. Furthermore, the inclusion of ordering features in the graph makes it easier to check how long the respective portion of the measurement data orthe respective processing product is still required for further processing.

[0024] Access via ordering attributes also has the advantage that each portion of the measurement data or each processing product is automatically accessible via a unique identifier. This eliminates the need to calculate a hash value, which is only available once the entire portion of the measurement data or the processing product is available. Furthermore, the graph structure also enables one-to-multiple relationships between ordering attributes.

[0025] In the use case mentioned above of measurement data collected during vehicle test drives, for example, nodes can reference projects currently being worked on. Each node corresponding to a project can then be connected, for example, via edges to other nodes corresponding to test drives. A test drive, the measurement data obtained from it, and the processing products generated from it can therefore be relevant to multiple projects. This can then be relevant, for example, for lifecycle rules that state that the original measurement data and at least deterministic instructions for generating all processing products must be retained until the final report for a specific project is approved.

[0026] In a path through the graph, ordering features can be mixed with designations of parts of the measured data and processing products. For example, the path / Project1 / Test Drive1 / Camera2 / Cutout[01:00-02:00] / Black and White

[0027] state that a specific minute is to be cut from the video data stream recorded by camera 2 during test drive 1 belonging to project 1 and converted to black and white. The temporal segment does not have to exist as a finished processing product, but can be created ad hoc.

[0028] Optionally, topology information can also be stored in the graph, which specifies a physical storage location for portions of measurement data and / or processing products. This makes it easier to perform further processing of the measurement data at its current storage location, instead of having to first transport the measurement data over a network. For example, the processing can be packaged in Docker or other containers that run in a standardized execution environment at the storage location of the respective data. This can particularly accelerate processing steps that use very large volumes of measurement data as source material and then heavily condense this data. For example, it is useful to create temporal segments of video data streams directly at the storage location of the original data, because then only these temporal segments need to be transported over the network, and not the complete original data.However, the original data is still accessible via the graph structure if needed again.

[0029] Furthermore, nodes in the graph can also be annotated with lifecycle policies, policies regarding the storage locations for the respective data, or even access rights. On a path through the graph leading to a specific node, one may encounter multiple policies, some of which complement each other and some of which contradict each other. In particular, it can then be generally defined, for example, how these different policies are to be offset against each other. If, for example, several policies specify different retention periods, the longest retention period can prevail. For example, the standard lifecycle policy for measurement data may state that the measurement data is to be retained for two weeks.However, if it turns out that a portion of the measurement data can be used as training data for a machine learning model and the trained machine learning model is incorporated into a specifically marketed product, the retention period for this portion of the measurement data can be extended to 30 years.

[0030] Alternatively, a specific policy can take precedence over a more general policy. In the example above, where the general retention period for measurement data is two weeks, a developer may determine that a certain portion of the measurement data is unusable for further work, for example, because the camera lens was dirty or the low sun caused the camera's image sensor to saturate. In this case, a specific policy can be set for the unusable portions of the measurement data, stating that they can be deleted after just one day.

[0031] The graph structure also facilitates the updating of processing products that depend on a roughly defined, but not definitive, set of measurement data portions or processing products. For example, a summary monthly report may depend on the results of all test drives conducted in the respective month. When measurement data from a new test drive is added, the monthly report node is connected via new edges to new nodes that represent the measurement data and, if applicable, any resulting processing products. Monitoring the monthly report node can react to this addition of new edges and trigger a regeneration of the monthly report based on the newly expanded set of connected nodes.Thus, it is generally advantageous to trigger an action with respect to a processing product represented by that node in response to the set of edges at a node changing. This action may, in particular, include, for example, • to immediately update the processing product and replace it with the updated version, and / or • to mark the previous version of the processing product as obsolete, but to retain it, and / or • to delete the processing product and only regenerate it when it is next accessed.

[0032] In particular, multiple versions of the processed product can coexist in the graph. For example, a version of the monthly report that has already been distributed to customers or other entities can remain unchanged, while a daily version is maintained at the same time.

[0033] The method can, in particular, be fully or partially computer-implemented. Therefore, the invention also relates to a computer program with machine-readable instructions that, when executed on one or more computers and / or compute instances, cause the computer(s) and / or compute instances to execute the described method. In this sense, control units for vehicles and embedded systems for technical devices that are also capable of executing machine-readable instructions are also to be regarded as computers. Compute instances can, for example, be virtual machines, containers, or serverless execution environments, which can be provided in a cloud, in particular.

[0034] The invention also relates to a machine-readable data carrier and / or a downloadable product containing the computer program. A downloadable product is a digital product that can be transmitted over a data network, i.e., downloaded by a user of the data network, and which can be offered for immediate download, for example, in an online shop.

[0035] Furthermore, one or more computers and / or compute instances may be equipped with the computer program, the machine-readable data carrier or the download product.

[0036] Further measures improving the invention are presented in more detail below together with the description of the preferred embodiments of the invention with reference to figures. Examples of implementation

[0037] It shows: Fig. 1 embodiment of the method 100 for storing measurement data 1 and processing products 2; Fig. 2 Example graph 3 generated using method 1.

[0038] Fig. 1 is a schematic flow diagram of an embodiment of the method 100 for storing measurement data 1 and processing products 2.

[0039] In step 110, a graph 3 is created. In this graph 3, nodes 4 correspond to the measurement data 1, to portions 1a of this measurement data 1, and / or to processing products of the measurement data 1 or portions 1a. Each node 4 corresponding to a specific processing product 2 is connected via edges 5 to those portions 1a of the measurement data 1, or to those processing products 2, from which this specific processing product 2 arises.

[0040] According to block 111, further nodes 4 and edges 5 of the graph 3 can represent ordering features for finding or accessing portions 1a of measurement data 1 or concrete processing products 2.

[0041] In step 120, in association with each node 4 corresponding to a concrete processing product 2 • this specific processing product 2, and / or • Instructions 2a for generating this specific processing product 2 from the measurement data 1, 1a or processing products 2 from which it arises are stored in a memory 6.

[0042] According to block 121, a data type 7 of the respective portion 1a of the measurement data 1 or of the respective processing product 2 can be stored in association with portions 1a of the measurement data 1 or with processing products 2.

[0043] According to block 122, a library 8 can additionally be provided in which, in association with a source data type 7a and • a target data type 7b and / or • a relation between the source data type 7a and the target data type 7b

[0044] Instructions 8a for converting data of source data type 7a into data of target data type 7b are stored.

[0045] According to block 123, at least one path 9 can be determined based on the graph 3 and / or using the library 8, on which a processing product 2 of a predetermined target data type 7b can be obtained starting from at least one predetermined portion 1a of the measurement data 1 and / or from at least one predetermined concrete processing product 2.

[0046] According to block 123a, a list of properties of the measurement data 1, 1a, or of the specific processing product 2 that are to be changed on the path to the target data type 7b can be determined from the specified portion (1a) of the measurement data 1 and / or from at least one specified concrete processing product 2, on the one hand, and the specified target data type 7b, on the other hand. A path 9 with several steps, in each of which a property contained in the list is changed, can then be determined according to block 123b.

[0047] According to block 123c, several candidate paths 9a can be determined. According to block 123d, these candidate paths 9a can each be assigned scores 9b with respect to a predetermined optimization objective. A candidate path 9a with the best score 9b can then be selected according to block 123e.

[0048] According to block 124, storage locations for measurement data 1, 1a and / or specific processing products 2 can be determined based on life cycle conditions and / or on data protection constraints for cross-border data exchange.

[0049] In the Fig. In the example shown in Figure 1, step 130 checks whether • a time elapsed since the creation of a specific processed product 2, and / or • an observed access behavior to this specific processing product 2 satisfies a predefined life cycle condition. If this is the case (truth value 1), in step 140, this specific processing product 2 is replaced by instructions 2a for generating this specific processing product 2 from the measurement data 1, 1a, or the processing products 2 from which it originates.

[0050] In the Fig. In the example shown in Figure 2, step 150 checks whether the quantity of processing products 2 generated from a specific portion 1a of the measurement data 1 or from a specific processing product 2 satisfies a predefined life cycle condition. If this is the case (truth value 1), this specific portion 1a of the measurement data 1 or this specific processing product 2 is deleted in step 160.

[0051] According to block 131 or 151, at least one lifecycle condition can be determined from a predefined list of queries that must be answered based on the measurement data 1, 1a, and the processing products 2. As previously explained, this "specification" of queries, specific to the application at hand, determines which data is required and for how long.

[0052] An exemplary graph 3 that can be generated within the framework of the method 100 is shown in Fig. 2. In this example, measurement data 1 comprises a video data stream recorded by a camera mounted on a vehicle. Of this, temporal segments 1a and 1a', which relate to specific events, are of particular interest.

[0053] The graph 3 thus has its root in a first node 4, which corresponds to the complete measurement data 1. From this, two edges 5 extend to further nodes 4, which correspond to the temporal segments 1a and 1a'. The temporal segments 1a and 1a' can also be understood as processing products 2 of the complete measurement data 1. For the sake of clarity, however, Fig. 2 distinguishes between the temporal segments 1a and 1a' on the one hand, and the processing products 2 created from them on the other hand.

[0054] In the Fig.In the example shown in Figure 2, the first temporal segment 1a is processed into a first processing product 2 based on a first instruction 2a, which corresponds to an edge 5. Similarly, the second temporal segment 1a' is processed into a second processing product 2' based on a second instruction 2a', which corresponds to another edge 5. Using a third instruction 2a'', which manifests itself in two further edges 5, the two processing products 2 and 2' are further processed into a third processing product 2''.

Claims

[1] Method (100) for storing measurement data (1) and processing products (2) generated therefrom, wherein a graph (3) is created (110) in which • Nodes (4) correspond to the measurement data (1) or parts (1a) of the measurement data (1), or to processing products (2), • each node (4) corresponding to a specific processing product (2) is connected via edges (5) to those parts (1a) of the measurement data (1) or to those processing products (2) from which this specific processing product (2) arises, and • in association with each node (4) corresponding to a specific processing product (2) ◯ this specific processing product (2), and / or ◯ Instructions (2a) for generating this specific processing product (2) from the measurement data (1, 1a) or processing products (2) from which it arises are stored (120) in a memory (6). [2] Method (100) according to claim 1, wherein in association with portions (1a) of the measurement data (1) or with processing products (2), a data type (7) of the respective portion (1a) of the measurement data (1) or of the respective processing product (2) is stored (121). [3] Method according to claim 2, wherein a library (8) is additionally provided (122) in which, in association with a source data type (7a) and • a target data type (7b) and / or • a relation between the source data type (7a) and the target data type (7b) contains instructions (8a) for converting data of the source data type (7a) into data of the target data type (7b). [4] Method (100) according to one of claims 2 to 3, wherein on the basis of the graph (3) and / or using the library (8), at least one path (9) is determined (123) on which, starting from at least one predetermined portion (1a) of the measurement data (1) and / or from at least one predetermined concrete processing product (2), a processing product (2) of a predetermined target data type (7b) can be obtained. [5] Method (100) according to claim 4, wherein • from the predetermined portion (1a) of the measurement data (1), and / or from at least one predetermined concrete processing product (2), on the one hand, and the predetermined target data type (7b), on the other hand, a list of properties of the measurement data (1, 1a) or of the concrete processing product (2) is determined (123a) which are to be changed on the way to the target data type (7b), and • a path (9) with several steps is determined (123b), in each of which a property contained in the list is changed. [6] Method (100) according to one of claims 4 to 5, wherein • several candidate paths (9a) are determined (123c), • these candidate paths (9a) are each assigned a rating number (9b) with regard to a given optimization goal (123d) and • a candidate path (9a) with the best score (9b) is selected (123e). [7] Method (100) according to one of claims 1 to 6, wherein at least one processing product (2) comprises a temporal section from the measurement data (1) and / or from a previous processing product (2). [8] Method (100) according to claim 7, wherein the temporal segment comprises a predetermined period of time before and after the time of an event identified in the measurement data (1). [9] Method (100) according to one of claims 1 to 8, wherein in response to • a time elapsed since the creation of a specific processed product (2), and / or • an observed access behavior to this concrete processing product (2) satisfies a predetermined life cycle condition (130), this concrete processing product (2) is replaced by instructions (2a) for generating this concrete processing product (2) from the measurement data (1, 1a) or processing products (2) from which it arises (140). [10] Method (100) according to one of claims 1 to 9, wherein in response to the quantity of processing products (2) generated from a specific portion (1a) of the measurement data (1), or from a specific processing product (2), satisfying a predetermined life cycle condition (150), this specific portion (1a) of the measurement data (1), or this specific processing product (2), is deleted (160). [11] Method (100) according to one of claims 9 to 10, wherein at least one life cycle condition is determined (131, 151) from a predetermined list of queries to be answered based on the measurement data (1, 1a) and the processing products (2). [12] Method according to one of claims 1 to 11, wherein storage locations for measurement data (1, 1a) and / or concrete processing products (2) are determined on the basis of life cycle conditions and / or on the basis of data protection boundary conditions for the cross-border data exchange (124). [13] Method according to one of claims 1 to 12, wherein further nodes (4) and edges (5) of the graph (3) represent ordering features for finding or accessing portions (1a) of measurement data (1) or concrete processing products (2) (111). [14] Method according to one of claims 1 to 13, wherein the measurement data (1) contain a temporal sequence of measured values ​​of at least one measured variable, and / or of point clouds and / or video frames whose point values ​​or pixel values ​​represent measured values ​​of at least one measured variable. [15] A computer program comprising machine-readable instructions which, when executed on one or more computers and / or compute instances, cause the computer(s) and / or compute instances to carry out the method (100) according to any one of claims 1 to 14. [16] Machine-readable data carrier and / or download product with the computer program according to claim 15.

Images