Apparatus, computer program, method for generating a pattern, and method for applying a pattern to an object

Abstract

Embodiments of the present disclosure relate to an apparatus, a computer program, a method for generating a pattern, and a method for applying a pattern to an object. The method for generating a pattern comprises obtaining at least one predefined constraint to the pattern and generating, using the computer vision algorithm, the pattern complying with the predefined constraint.

Description

[0001] Apparatus, computer program, method for generating a pattern, and method for applying a pattern to an object

[0002] Field

[0003] Embodiments of the present disclosure relate to an apparatus, a computer program, a method for generating a pattern, and a method for applying a pattern to an object. In particular, embodiments of the present disclosure relate to a concept for generating a pattern which is salient for a computer vision algorithm in an environment of an object provided with the pattern.

[0004] Background

[0005] High-visibility clothing is worn to increase a person’s visibility in dangerous situations. They can be used, for example, by a person involved in an accident on a motorway, by children near a school or by a cyclist who wants to be seen by drivers. They are designed to be highly luminescent, making them easily discernible from any background. However, they are intended specifically for human vision. Autonomous vehicles, also known as self-driving cars, are becoming more and more efficient and are already used in many specific situations. Although many types of sensors can be used, cameras are the most common. Artificial intelligence is used to analyze images of the vehicle's environment and to extract important information that will be used by the algorithm to take a decision. Although enormous efforts are being made to make artificial intelligence models more and more efficient, they may fail to detect something important (such as a cyclist or a child). Hence, traffic participants wearing conventional high-visibility clothing may be missed by an autonomously (or at least partly autonomously) driving vehicle.

[0006] Apart from this, features of high-visibility clothing may impair its visibility to computer vision algorithms. Hence, there is a demand for an improved concept of providing a pattern salient for a computer vision algorithm. Summary

[0007] This demand may be satisfied by the subject-matter of the appended independent and dependent claims.

[0008] Embodiments of the present disclosure provide a method for generating a pattern to detect an object provided with the pattern by a computer vision algorithm. The method comprises obtaining at least one predefined constraint to the pattern and generating, using the computer vision algorithm, the pattern complying with the predefined constraint. In this way, predefined features can be considered in generating the pattern to reduce a degradation of a visibility of the pattern through the features to the computer vision algorithm. In practice, the proposed method thus allows to consider features that, e.g., improve a visibility to persons.

[0009] In various examples, generating the pattern comprises obtaining an initial pattern, obtaining at least one image of a sample of the object, applying the initial pattern to the sample in the image to provide at least one sample image, applying the computer vision algorithm to the sample image to determine an accuracy of a characterization of the sample provided with the initial pattern by the computer vision algorithm, and adapting the initial pattern to improve the accuracy for generating the pattern for the object.

[0010] A skilled person having benefit from the present disclosure will appreciate that the constraint may relate to any desired feature of the pattern. The constraint may provide for a predefined color, a predetermined sign, and / or a predefined contrast in the pattern.

[0011] The initial pattern may be adapted in consideration of the computer vision algorithm. In doing so, e.g., an underlying model of the computer vision algorithm is considered. This allows for an algorithm-specific pattern which is more reliably detected by the computer vision algorithm. Also, it allows for a more efficient and faster generation of the pattern.

[0012] Alternatively, the initial pattern may be adapted irrespective of the computer vision algorithm. In this case, the initial pattern is adapted based on an algorithm-unspecific approach. This, e.g., may allow for a more algorithm-unspecific pattern that is reliably detected by various computer vision algorithms. In some embodiments, the method further may comprise providing the generated pattern to apply the generated pattern to the object.

[0013] In various examples, the object may comprise a garment, a traffic sign, and / or a patch therefor.

[0014] For example, the garment may comprise a high-visibility jacket.

[0015] In some examples, the method further may comprise obtaining a reversed adversarial attack algorithm, the reversed adversarial attack algorithm may be derived from an adversarial attack algorithm for generating adversarial examples for the computer vision algorithm and configured to adapt the initial pattern to improve the accuracy, and adapting the initial pattern may comprise adapting the initial pattern using the reversed adversarial attack algorithm. In this way, benefit is made from a vulnerability of neural networks to adversarial attacks. Adversarial attacks are suitable for various computer vision algorithms or systems because they are related to the tasks rather than to models underlying the algorithms (e.g., two models trained to solve the same task will be vulnerable to the same adversarial examples of the same adversarial attack algorithm). Accordingly, the generated pattern is algorithm-unspecific and, so, more visible to various computer vision algorithms. This is an important advantage over other existing approaches where generated patterns are strongly linked to the model.

[0016] A skilled person having benefit from the present disclosure will appreciate that adversarial attack algorithms in practice do not require training data or equivalent information on the computer vision algorithm to generate well-functioning adversarial examples. Likewise, the use of the reversed adversarial attack algorithm removes the requirement for the training data of the computer vision algorithm to generate the pattern. Hence, the proposed approach allows to generate salient patterns for unknown computer vision algorithms.

[0017] In practice, the proposed concept may be applied for any type of computer vision algorithms. In some examples, the computer vision algorithm may be configured for the use in automotive applications, traffic infrastructure, and / or robot applications. For example, the computer vision algorithm may be configured for object detection and / or object recognition. Other embodiments relate to a method for producing an object and for applying a pattern to an object, the method comprising obtaining a pattern which is obtainable by a method according to any one of the preceding claims, and applying the pattern to the object.

[0018] The skilled person will appreciate that the proposed concept applies for various applications of the pattern. In some examples, the object may comprise a garment, a traffic sign, and / or a patch therefor. For example, the garment may comprise a hi gh-visibility jacket

[0019] An aspect of the present disclosure relates to a computer program having a program code for performing a method according to the present disclosure when the program may be executed on a processor or a programmable hardware.

[0020] Some aspects of the present disclosure relate to an apparatus comprising one or more interfaces, and a processing circuit configured to control the one or more interfaces and to carry out a method according to the present disclosure using the one or more interfaces.

[0021] Brief description of the Figures

[0022] Some examples of apparatuses and / or methods will be described in the following by way of example only, and with reference to the accompanying figures, in which

[0023] Fig. 1 shows a flow chart schematically illustrating an embodiment of a method for generating a pattern to detect an object provided with the pattern by a computer vision algorithm;

[0024] Fig. 2a and 2b show a flow chart schematically illustrating another embodiment of the method;

[0025] Fig. 3 shows an exemplary use case of an embodiment of the proposed approach; and

[0026] Fig. 4 shows a flow chart schematically illustrating an embodiment of a method producing an object and for applying a pattern to an object. Detailed Description

[0027] Some examples are now described in more detail with reference to the enclosed figures. However, other possible examples are not limited to the features of these embodiments described in detail. Other examples may include modifications of the features as well as equivalents and alternatives to the features. Furthermore, the terminology used herein to describe certain examples should not be restrictive of further possible examples.

[0028] Throughout the description of the figures same or similar reference numerals refer to same or similar elements and / or features, which may be identical or implemented in a modified form while providing the same or a similar function. The thickness of lines, layers and / or areas in the figures may also be exaggerated for clarification.

[0029] When two elements A and B are combined using an “or”, this is to be understood as disclosing all possible combinations, i.e., only A, only B as well as A and B, unless expressly defined otherwise in the individual case. As an alternative wording for the same combinations, "at least one of A and B" or "A and / or B" may be used. This applies equivalently to combinations of more than two elements.

[0030] If a singular form, such as “a”, “an” and “the” is used and the use of only a single element is not defined as mandatory either explicitly or implicitly, further examples may also use several elements to implement the same function. If a function is described below as implemented using multiple elements, further examples may implement the same function using a single element or a single processing entity. It is further understood that the terms "include", "including", "comprise" and / or "comprising", when used, describe the presence of the specified features, integers, steps, operations, processes, elements, components and / or a group thereof, but do not exclude the presence or addition of one or more other features, integers, steps, operations, processes, elements, components and / or a group thereof.

[0031] In practice, computer vision algorithms for (partly) autonomously driving vehicles may struggle detecting high-visibility jackets. So far, it has been tried to further improve detection algorithms to improve a characterization of an environment of such vehicles. The present disclosure proposes an innovative approach which, instead of improving the detection algorithm, suggests developing high-visibility assets (e.g., a jacket) that will be more easily detected by computer vision algorithms. Further, the approach proposed herein allows the consideration of further demands on the pattern, e.g., sufficient visibility to humans and / or predefined features. Further details of the proposed approach are specified below with reference to Fig. 1.

[0032] Fig. 1 shows a flow chart schematically illustrating an embodiment of a method 100 for generating a pattern to detect an object provided with the pattern by a computer vision algorithm. Method 100 may be used to improve the detection of the object. The pattern can be understood as any combination of shapes, colors, and or contrasts. For example, the pattern is a drawing, an image, or any digital representation thereof. In practice, the pattern may include a combination or structure of one or more different shapes, figures, lines, and / or colors. As well, the pattern may be a closed / limited or a continuous pattern, i.e., a pattern in predefined boundaries or a pattern that may be continued or repeated arbitrarily often. A skilled person having benefit from the present disclosure will appreciate that the proposed concept may be applied for one or more dimensions. So, the pattern may be a one-, two-, or three-dimensional pattern. In various applications, the pattern is a two-dimensional pattern,

[0033] The object can be any (potential) target of the computer vision algorithm. In automotive applications, the object can be any object that occurs in traffic, e.g., vehicles, traffic signs, and / or garment for pedestrians (e.g., a high-visibility jacket or other garment for pedestrians). As well, the object may be of any other type of targets which is supposed to be detected and / or classified by a computer vision algorithm. The computer vision algorithm can be any type of computer vision algorithm, e.g., an object detection and / or object recognition algorithm. In practice, the computer vision algorithm comprises or utilizes a (trained) neural network.

[0034] As can be seen from the flow chart, method 100 comprises obtaining 110 at least one predefined constraint to the pattern. The constraint may relate to a desired feature or property the generated pattern should have. The desired feature or property, e.g., provides a sufficient or a favorable visibility of the pattern by humans. For this, the pattern is supposed to exhibit a certain color, contrast, and / or any other feature which may be determined manually (i.e., by a user) or automatically (i.e., by an algorithm). In applications, e.g., a human and / or an algorithm automatically determines a certain combination of colors, a contrast, and / or parts of the pattern (from experimental data) for a favorable visibility of the pattern by humans. Accordingly, the constraint, e.g., is obtained from a user input or another algorithm configured to determine the constraint. The property or feature, e.g., comprises a predefined color, contrast of colors, and / or reflective areas or stripes for a sufficient / desired visibility by humans. Also, the constraint may comprise a (legal) requirement of the pattern. As well, the property or feature may comprise text or a sign, e.g., a warning message or sign. In this way, the pattern is more informative to humans. In practice, the text or sign may also comprise a design element, e.g., a logo or a brand name.

[0035] Further, method 100 comprises generating 120, using the computer vision algorithm, the pattern complying with the predefined constraint. In doing so, the pattern is adapted such that the pattern both satisfies the constraint and is more reliably and / or more accurately recognized and / or characterized by the computer vision algorithm. In order to do so, an (in)accuracy, the so-called “loss function” may be considered. By considering the constraint in generating the pattern, it can be ensured that the pattern exhibits the desired feature or property and that the pattern is salient to the computer vision algorithm. An advantage of the present approach is that this way, the desired feature or property is already considered in generating the pattern for a favorable or sufficient visibility to or detection / recognition by the computer vision algorithm. Thus, the generated pattern may be more visible and / or more easily identifiable by the computer vision algorithm than in concepts where a desired feature or property is added afterwards, i.e., after generating the pattern.

[0036] So, the pattern may be created so that two or more objectives are considered at the same time: In various use cases, the pattern should improve the accuracy of a computer vision algorithm, such as an object detector, to make it more visible for it, and secondly, the pattern should be sufficiently visible to humans. So, the pattern will therefore be more visible to both humans and a computer vision algorithm. To do this, various approaches can be applied of which only few are described herein in more detail. In some embodiments, the pattern is initialized with properties that enhance visibility for humans (e.g. flash colors and high contrast). Then, during pattern optimization, constraints will be applied to force the pattern to retain these properties.

[0037] More details and aspects regarding the generation of the pattern are described with reference to Fig. 2a and 2b.

[0038] Fig. 2a shows a flow chart of a pattern generation process. Fig. 2b shows an exemplary use case. As can be seen from picture “A” (left) of Fig. 2b, the use case, e.g., relates to a high-visibility jacket 270, for instance a high-visibility jacket for persons in traffic. For a sufficient visibility to other persons in traffic, the jacket 270 exhibits a (fluorescent) signal color, here neon yellow, and reflective areas 272 for an even higher visibility if the jacket 270 is illuminated. Even though the jacket 270 may be sufficiently visible to humans, sensors and / or computer vision algorithms of (partly) autonomously driving vehicles may fail to recognize and / or classify the jacket 270. In the present example, the computer vision algorithm, e.g., recognizes persons wearing the high-visibility jacket only with a probability of 57% (see picture “A”). Hence, the proposed approach of the present disclosure, e.g., may be applied to generate a pattern for a high-visibility jackets.

[0039] As can be seen from the flow chart of Fig. 2a, in order to do so, a pattern is initialized either randomly or with constraints. For this, in step 210 of the procedure, an initial pattern is obtained. In general, the initial pattern can be any random or predefined pattern with arbitrary shapes and colors. In a following step 230, transformations are applied to the pattern to create more diversity and make it more "real" when combined with an image for more realistic samples in the further procedure. The transformations are done to minimize the loss function (e.g. optimize the detection).

[0040] In another step 220 one or more images of a sample of the object are obtained. For this, e.g., a dataset containing real world images is received. In practice, the images may not only indicate a single object but various objects. In the present use case, the images, e.g., show various exemplary use cases of high-visibility jackets, e.g., of persons who could wear a high-visibility jacket, e.g., cyclists or pedestrians. The skilled person will appreciate that the images do not need to be the images which were used to train the computer vision algorithm but may be also any other images.

[0041] In a further step 240, the images are combined with the pattern, i.e., the initial pattern is applied to the sample in the images to provide sample images. In doing so, e.g., the pattern is added to the persons. As can be seen from picture “B” of Fig. 2b, for a high-visibility jacket, the pattern is inserted in the images such that it appears as if the persons in the images wear a high-visibility jacket 280 provided with the pattern. In practice, this may be done automatically, e.g., by a suitable image manipulation algorithm, and / or the pattern is applied manually to the images. In order to do that automatically, the computer vision algorithm can automatically extract positions of people. In various applications, the pattern may be applied partly manually and partly automatically, e.g., through automatic extraction of the persons and manual verification of the extracted persons or their bounding boxes.

[0042] As can be seen from picture “B”, in the present case, the pattern is subject to a constraint, here, e.g., reflective areas 282 in the pattern for a higher visibility to humans when the high visibility jacket 280 is illuminated, e.g., by headlights.

[0043] Then, in another step 250, the computer vision algorithm is applied to the sample images to determine an accuracy of a characterization of the sample provided with the initial pattern by the computer vision algorithm. The characterization, e.g., provides for detection, recognition, and / or classification of objects / persons in the images. In the present case, the computer vision algorithm is configured to recognize persons wearing a high-visibility jacket.

[0044] To evaluate the characterization, a loss function is applied to the characterization of the computer vision algorithm in a further step 260. The loss function indicates how accurate / inaccu- rate or reliable the characterization is. In the present case, the loss function, e.g., indicates how reliably a person wearing the high-visibility jacket 280 is recognized in the images by the computer vision algorithm.

[0045] Then, the loss function is used to update the pattern by modifying the pattern in a way that the characterization is improved, i.e., that the characterization is more accurate, thereby generating an adapted pattern 280’, as shown in picture “C”. In the present case, the loss function, e.g., indicates how reliably or how accurate persons are recognized in the images combined with the pattern. The pattern, then, may be adapted such that the loss function decreases or, ideally, becomes minimal.

[0046] In some embodiments, the pattern can be adapted in consideration of the computer vision algorithm, e.g., by backpropagation through the model (“White box optimization”), if the underlying model is accessible or available. In doing so, the pattern is specifically adapted such that the loss function will decrease in consideration of the underlying model. Otherwise, if the model is not accessible or not available, e.g., because access to a program code of the computer vision algorithm is limited or not available (e.g., for ownership reasons), a so-called "black box" optimization may be applied.

[0047] The skilled person will appreciate that for white box optimization as well as for black box optimization various optimization approaches can be used and that the optimization may be repeated until a desired or favorable level of accuracy or reliability is achieved.

[0048] When adapting the pattern, it is provided that the pattern still fulfills the constraint - in the present case, that the reflective areas are preserved. For this, the adaptions provide for only changing features or properties of the pattern except for the reflective areas. To this end, the optimization process may be adapted such that it considers the constraint. So, in the present example the optimization process can be adapted such that it keeps the reflective areas and that it only changes other features or properties (if necessary) to improve the visibility by the computer vision algorithm.

[0049] In order to do so, the pattern may be initialized with a certain desired or predefined feature or property, here, e.g., the reflective areas 282, and certain conservation constraints (e.g. the reflective areas cannot be changed too much. E.g. a size of the reflective areas 282 may not be changed by more than a predefined value) may be applied in adapting the pattern 280, we create a high visibility pattern around a specific area.

[0050] In this way, it may be ensured that not only the computer vision algorithm (of an autonomously driving vehicle) but also humans may reliably recognize persons wearing a high visibility jacket provided with the pattern. To provide a certain recognition reliability, i.e., a probability that the computer vision algorithm recognizes a person wearing the high visibility jacket, the pattern can be adapted repeatedly in the proposed way, i.e., such that the adapted pattern 280’ still fulfills the constraint. In this way, the pattern is adapted until a recognition reliability is achieved. In the present examples, the pattern is modified until a recognition reliability of 99% is reached to obtain a final (optimized) pattern.

[0051] The final pattern, then, may be applied to a high visibility jacket. In practice, the pattern, e.g., is printed or otherwise applied to the high visibility jacket. It is noted that the proposed approach does not only apply for high visibility jackets but also for any other potential target of a computer vision algorithm. In practice, the pattern, e.g., may be applied to any other equipment or gear (of traffic participants) (e.g., helmets, gloves, bicycles, warning triangle), to any infrastructure element (e.g., road signs, bridges, traffic lights, obstacles), to vehicles, and / or any other object in traffic.

[0052] Further, the proposed approach applies for various constraints. In practice, the constraint may not only apply to features or properties for better perception by humans or person but also for informative features such as warning messages, traffic signs, design elements, and / or brand logos. So, in some implementations, the constraint, e.g., provides for a traffic sign or other informative elements in the pattern. In various embodiments, the constraint provides for a predefined color, a predetermined sign, and / or a predefined contrast in the pattern. As can be seen from Fig. 3, the constraint, e.g., provides for a brand logo 283 in the pattern or on the object.

[0053] It is noted that the proposed approach of increasing the visibility or identifiability of the pattern is not limited only to high visibility clothing for autonomous vehicles. Objects can also be optimized to be more detectable by autonomous vehicles, such as road signs, traffic lights or, construction vehicles working on the motorway (which are more likely to be undetected). Also, the system is not only limited to autonomous vehicles. It can be applied in other situations that require detection. The high visibility patches can be applied to workers or objects in factories that use robots (such as in a warehouse), or it can also be applied to food packaging, as more and more shops are using object detection at the checkout. All detection systems where the physical environment can be modified could benefit from the generation of patterns according to the present disclosure. So, in general, the subject-matter of the present disclosure is applicable to various objects. In practice, the object, e.g., comprises or corresponds to equipment of a traffic participant, a garment, a traffic sign, and / or a patch therefor. The skilled person will appreciate that the present approach also applies to any target object in non-ve- hicular applications.

[0054] In some embodiments of the above adaption is carried out using an adversarial attack algorithm adapted to improve a visibility or recognizability of a pattern for a computer vision algorithm, herein referred to as “reversed adversarial attack algorithm”. The reversed adversarial attack algorithm is derived from an adversarial attack algorithm for generating adversarial examples for the computer vision algorithm and configured to adapt the initial pattern to improve the accuracy, and wherein adapting the initial pattern comprises adapting the initial pattern using the reversed adversarial attack algorithm.

[0055] In practice, adversarial attack algorithms are used to fool computer vision algorithms. For this, adversarial attack algorithms, e.g., create malicious images called adversarial examples which lead to false detections and / or classifications. Adversarial attack algorithms are based on the finding that a small, specifically designed perturbation in an input image can trick the artificial vision system into making errors. Adversarial attacks have been pushed even further with the creation of ‘physical’ adversarial examples where it is not necessary to directly modify the input image but rather the real world (e.g., by adding stickers). In this way, e.g., computer vision algorithms for vehicles, e.g., (partly-) autonomously driving vehicles, face recognition, or other purposes can be fooled. A strength of adversarial attacks is that they do not need access to the system under attack and they work against different detection algorithms (for example, different car manufacturers). The skilled person will appreciate that the concept of adversarial attack algorithms can be also applied to the present proposed approach. In doing so, certain adjustments that change the goal of the adversarial attack algorithm from creating adversarial attack samples for fooling a computer vision algorithm into generating a pattern which is salient to the computer vision algorithm may be applied. In this way, the proposed approach uses the strength of adversarial attacks but reverses the disruptive approach to improve the visibility or identifiability of the pattern to the computer vision algorithm instead of fooling the computer vision algorithm.

[0056] As adversarial examples can be imperceptible to the human eyes, the initial pattern (e.g. a logo) can remain seemingly unchanged even after applying the reversed adversarial attack algorithm for adapting the pattern for increasing its visibility / identifiability to the computer vision algorithm.

[0057] In practice, a method for generating the pattern may be implemented in connection with a method for applying the generated pattern. Fig. 4 exemplarily shows an embodiment of a method 400 for producing an object and for applying a pattern to an object. Method 400 comprises obtaining 410 a pattern which is obtainable by a method proposed herein for generating a pattern, e.g., method 100. For this, the pattern, e.g., is obtained from a memory or data processing circuit providing the pattern generated by the proposed method. Further, method 400 comprises applying 420 the pattern to the object. To this end, the pattern, e.g., is printed onto the object or otherwise applied (e.g., embroidered). Also, the pattern may be applied to a patch or a sticker that can be applied to the object. For this, a suitable printing device may be used.

[0058] The present approach can be also implemented in an apparatus, as explained in more detail below with reference to Fig. 5.

[0059] Fig. 5 shows a block diagram schematically illustrating an apparatus 500 comprising one or more interfaces 510 and a processing circuit 520 configured to control the one or more interfaces and to carry out a method of the present disclosure using the one or more interfaces 510.

[0060] Examples of the interfaces 510 comprise wired or wireless interfaces. In general, the interfaces can comprise any means for communicating (and optionally processing) signals, here, e.g., for examples for the purpose of communicating data for generating the pattern.

[0061] The processing circuit can be any data processing circuit for executing any one of the proposed methods. In embodiments, the processing circuit can comprise any programmable hardware (microcontroller, field-programmable-gate array, FPGA, central processing unit, CPU, graphics processing unit, GPU, or the like).

[0062] As the skilled person will understand, features and aspects of the proposed methods may also apply to the proposed apparatus. Accordingly, features and aspects of the methods may be also implemented, mutatis mutandis, in the proposed apparatus.

[0063] Further embodiments pertain to:

[0064] (1) A method for generating a pattern to detect an object provided with the pattern by a computer vision algorithm, the method comprising: obtaining at least one predefined constraint to the pattern; generating, using the computer vision algorithm, the pattern complying with the predefined constraint.

[0065] (2) The method of (1), wherein generating the pattern comprises: obtaining an initial pattern; obtaining at least one image of a sample of the object; applying the initial pattern to the sample in the image to provide at least one sample image; applying the computer vision algorithm to the sample image to determine an accuracy of a characterization of the sample provided with the initial pattern by the computer vision algorithm; and adapting the initial pattern to improve the accuracy for generating the pattern for the object.

[0066] (3) The method of (2), wherein the method further comprises obtaining a reversed adversarial attack algorithm, wherein the reversed adversarial attack algorithm is derived from an adversarial attack algorithm for generating adversarial examples for the computer vision algorithm and configured to adapt the initial pattern to improve the accuracy, and wherein adapting the initial pattern comprises adapting the initial pattern using the reversed adversarial attack algorithm.

[0067] (4) The method of any one of (1) to (3), wherein the constraint provides for a predefined color, a predetermined sign, and / or a predefined contrast in the pattern.

[0068] (5) The method of any one of (1) to (4), wherein the initial pattern is adapted in consideration of the computer vision algorithm or irrespective of the computer vision algorithm. (6) The method of any one of (1) to (5), wherein the method further comprises providing the generated pattern to apply the generated pattern to the object.

[0069] (7) The method of any one of (1) to (6), wherein the object comprises a garment, a traffic sign, and / or a patch therefor.

[0070] (8) The method of (7), wherein the garment comprises a high-visibility j acket.

[0071] (9) The method of any one of (1) to (8), wherein the computer vision algorithm is configured for the use in automotive applications, traffic infrastructure, and / or robot applications.

[0072] (10) The method of any one of (1) to (9), wherein the computer vision algorithm is configured for object detection and / or object recognition.

[0073] (11) A method for producing an object and for applying a pattern to an object, the method comprising: obtaining a pattern which is obtainable by a method according to any one of the preceding claims; and applying the pattern to the object.

[0074] (12) The method of (11), wherein the object comprises a garment, a traffic sign, and / or a patch therefor.

[0075] (13) The method of (12), wherein the garment comprises a high-visibility jacket

[0076] (14) A computer program having a program code for performing a method according to any one of (1) to (12) when the program is executed on a processor or a programmable hardware.

[0077] (15) An apparatus comprising: one or more interfaces; and a processing circuit configured to control the one or more interfaces and to carry out a method of any one of (1) to (12) using the one or more interfaces.

[0078] The aspects and features described in relation to a particular one of the previous examples may also be combined with one or more of the further examples to replace an identical or similar feature of that further example or to additionally introduce the features into the further example.

[0079] Examples may further be or relate to a (computer) program including a program code to execute one or more of the above methods when the program is executed on a computer, processor or other programmable hardware component. Thus, steps, operations or processes of different ones of the methods described above may also be executed by programmed computers, processors or other programmable hardware components. Examples may also cover program storage devices, such as digital data storage media, which are machine-, processor- or computer-readable and encode and / or contain machine-executable, processor-executable or computer-executable programs and instructions. Program storage devices may include or be digital storage devices, magnetic storage media such as magnetic disks and magnetic tapes, hard disk drives, or optically readable digital data storage media, for example. Other examples may also include computers, processors, control units, (field) programmable logic arrays ((F)PLAs), (field) programmable gate arrays ((F)PGAs), graphics processor units (GPU), application-specific integrated circuits (ASICs), integrated circuits (ICs) or system-on-a-chip (SoCs) systems programmed to execute the steps of the methods described above.

[0080] It is further understood that the disclosure of several steps, processes, operations, or functions disclosed in the description or claims shall not be construed to imply that these operations are necessarily dependent on the order described, unless explicitly stated in the individual case or necessary for technical reasons. Therefore, the previous description does not limit the execution of several steps or functions to a certain order. Furthermore, in further examples, a single step, function, process, or operation may include and / or be broken up into several sub-steps, -functions, -processes or -operations. If some aspects have been described in relation to a device or system, these aspects should also be understood as a description of the corresponding method. For example, a block, device or functional aspect of the device or system may correspond to a feature, such as a method step, of the corresponding method. Accordingly, aspects described in relation to a method shall also be understood as a description of a corresponding block, a corresponding element, a property or a functional feature of a corresponding device or a corresponding system.

[0081] The following claims are hereby incorporated in the detailed description, wherein each claim may stand on its own as a separate example. It should also be noted that although in the claims a dependent claim refers to a particular combination with one or more other claims, other examples may also include a combination of the dependent claim with the subject matter of any other dependent or independent claim. Such combinations are hereby explicitly proposed, unless it is stated in the individual case that a particular combination is not intended. Furthermore, features of a claim should also be included for any other independent claim, even if that claim is not directly defined as dependent on that other independent claim.

Claims

Claims1. A method for generating a pattern to detect an object provided with the pattern by a computer vision algorithm, the method comprising: obtaining at least one predefined constraint to the pattern; generating, using the computer vision algorithm, the pattern complying with the predefined constraint.

2. The method of claim 1, wherein generating the pattern comprises: obtaining an initial pattern; obtaining at least one image of a sample of the object; applying the initial pattern to the sample in the image to provide at least one sample image; applying the computer vision algorithm to the sample image to determine an accuracy of a characterization of the sample provided with the initial pattern by the computer vision algorithm; and adapting the initial pattern to improve the accuracy for generating the pattern for the object.

3. The method of claim 2, wherein the method further comprises obtaining a reversed adversarial attack algorithm, wherein the reversed adversarial attack algorithm is derived from an adversarial attack algorithm for generating adversarial examples for the computer vision algorithm and configured to adapt the initial pattern to improve the accuracy, and wherein adapting the initial pattern comprises adapting the initial pattern using the reversed adversarial attack algorithm.

4. The method of claim 1, wherein the constraint provides for a predefined color, a predetermined sign, and / or a predefined contrast in the pattern.

5. The method of claim 4, wherein the initial pattern is adapted in consideration of the computer vision algorithm or irrespective of the computer vision algorithm.

6. The method of claim 1, wherein the method further comprises providing the generated pattern to apply the generated pattern to the object.

7. The method of claim 1, wherein the object comprises a garment, a traffic sign, and / or a patch therefor.

8. The method of claim 7, wherein the garment comprises a high-visibility jacket.

9. The method of claim 1, wherein the computer vision algorithm is configured for the use in automotive applications, traffic infrastructure, and / or robot applications.

10. The method of claim 1, wherein the computer vision algorithm is configured for object detection and / or object recognition.

11. A method for producing an object and for applying a pattern to an object, the method comprising: obtaining a pattern which is obtainable by a method according to any one of the preceding claims; and applying the pattern to the object.

12. The method of claim 11, wherein the object comprises a garment, a traffic sign, and / or a patch therefor.

13. The method of claim 12, wherein the garment comprises a high-visibility jacket14. A computer program having a program code for performing a method according to claim 1 when the program is executed on a processor or a programmable hardware.

15. An apparatus comprising: one or more interfaces; and a processing circuit configured to control the one or more interfaces and to carry out a method of claim 1 using the one or more interfaces.

Images