Method of providing calibration data, method and apparatus for manufacturing a predefined point-symmetrical area

Abstract

The present invention relates to a method for providing calibration data (135) to calibrate a camera (102). The method includes the step of reading image data (105) provided by means of the camera (102). The image data (105) represents a camera image of at least one predefined point symmetry region (110). The method further includes the steps of using the image data (105) and determination rules (128) to determine at least one center of symmetry (112) of the at least one point symmetry region (110); comparing the position of the center of symmetry (112) in the camera image with a predefined position of a reference center of symmetry in a reference image (115) to determine a positional deviation (131) between the center of symmetry (112) and the reference center of symmetry; and using the positional deviation (131) to obtain displacement information (133) of at least one subset of pixels in the camera image relative to corresponding pixels in the reference image (115). The displacement information (133) is used to provide the calibration data (135).

Description

Technical Field

[0001] This invention is based on the apparatus or method of the type claimed in the independent claims. Computer programs are also the subject of this invention. Background Technology

[0002] In the field of camera calibration, targets with simple black-and-white patterns, mostly constructed from squares or circles, can be used. In particular, checkerboard patterns or regularly arranged circles can often be used, such as black circles arranged in a grid pattern on a white background, or vice versa.

[0003] DE 10 2020 202 160 A1, which will be published later, discloses a method for determining symmetry properties in image data and a method for controlling functions. Summary of the Invention

[0004] Against this backdrop, the proposed solution here provides a method according to the main claim, an apparatus for using the method, and a corresponding computer program. Advantageous extensions and improvements can be made to the apparatus described in the independent claim through the measures listed in the dependent claims.

[0005] According to the implementation method, in particular, the fact that points or objects in the world are marked or will be marked by means of point-symmetric regions can be utilized, enabling systems with imaging sensors and suitable methods proposed herein to detect and locate these point-symmetric regions with high precision in order to robustly and locally perform specific technical functions, optionally without being perceived as interference by humans or organisms.

[0006] For example, a symmetrical region might not be fully imaged into the camera image, for instance, because it might be partially occluded by an object, partially protrude from the image, or be cropped. Advantageously, the accuracy of the point's center of symmetry can still be maintained because partial occlusion does not distort its position: the remaining point symmetry pairs can still vote for the correct center of symmetry. Partial occlusion only reduces the strength of clustered points in the voting matrix, but the position of the center of symmetry can be preserved and still determined accurately and easily. This is a special advantage of utilizing point symmetry.

[0007] Further advantages in finding regions or patterns based on point symmetry can stem particularly from the fact that point symmetry is invariant with respect to rotation between the point-symmetric region and the camera or image recording, and is largely invariant with respect to the viewing angle. For example, the point-symmetric plane can be invariant with respect to affine imaging. The imaging of an arbitrarily oriented plane by a real camera can always be approximated well, at least locally, by affine imaging. For example, if a circular point-symmetric region is viewed from an oblique angle, the circle becomes an ellipse, in which the point symmetry characteristics and the center of point symmetry are preserved. Therefore, at least one point-symmetric region does not necessarily have to be viewed from a frontal view—even a very oblique viewpoint will not pose a difficulty, and achievable accuracy can be maintained. This invariance, especially with respect to rotation and with respect to the viewing angle, eliminates the need for precautions such as properly aligning the camera with the symmetric region or vice versa. Rather, it may be sufficient that the corresponding point-symmetric region is at least partially captured in the camera image, making it detectable. The relative positional relationship or arrangement between the point-symmetric region and the camera can be insignificant or almost insignificant in this case.

[0008] Regarding camera calibration, it is advantageous, particularly, to utilize a truly two-dimensional rather than one-dimensional surface that can be used for at least one point-symmetric region. Unlike, for example, conventional black-and-white calibration targets, the point-symmetric region can contain more information because the two-dimensional surface of the point-symmetric region can be largely used to accommodate the information therein. For conventional black-and-white calibration targets, the information they contain is mostly one-dimensional in nature because the information is particularly crammed into lines generated by the encounter of black and white surfaces. In the case of such a checkerboard pattern, this results in straight lines whose intersections represent so-called marker points that can be used for calibration. In the case of a circular surface as the target, circles are generated at the edges, and the centers of these circles form the marker points. Therefore, only (one-dimensional) lines contribute to the marker points. In other words, the surface of the target is utilized only to a reduced extent. In contrast, according to the implementation, particularly according to the design of the region or pattern, provided that a symmetric partner has been found, every point on the entire surface can contribute to determining the center of symmetry. The center of symmetry here refers to the marker point in the aforementioned sense. The entire surface can contribute to this. The more point pairs contribute to the formation of the center of symmetry, the stronger and more defined the center of symmetry can be imaged in the voting matrix, for example, thus enabling reliable detection and a low risk of false detection. Furthermore, the center of symmetry can be located more accurately (with high precision in the sub-pixel range) because it stands out more strongly from noise. In other words, the proposed symmetric regions and / or patterns can truly utilize available surfaces in a two-dimensional manner, translating into advantages in accuracy and robustness. Avoiding extreme areas of the camera and HDR effects can also be advantageous. Unlike traditional work using black-and-white patterns, working at extreme points can be avoided in terms of camera characteristic curves or camera signals that depend on the brightness at the corresponding pixels. Working at extreme points would be disadvantageous because camera characteristics differ significantly between dark and bright regions.

[0009] Below, we clarify the above facts based on some simple considerations. Light is composed of photons. From a camera's perspective, black means no photons, while in the case of white, each surface has many photons. A pixel (in the sense of the photosensitive element of a camera sensor) can only hold a limited number of photons (saturation capacity). This can lead to nonlinear effects, or even saturation effects, in bright areas. The noise of modern image sensors is almost entirely determined by the number of photons arriving at the pixel. In dark areas (fewer photons), the absolute noise may be lower, but the relative noise (related to brightness) may be higher. In bright areas (many photons), the absolute noise may be higher, but the relative noise may be lower. In this respect, black and white image areas may behave differently. Stray light effects in the optical path can cause bright image areas to spill into adjacent dark image areas, but not vice versa. Blur effects can also occur due to poor lens characteristics or misfocusing. Here, bright areas can also be blurred into dark areas, but not vice versa. The combination of blur (for whatever reason) and saturation in bright areas can cause bright saturated areas to spill into dark areas, making it impossible to find markers in the case of a traditional checkerboard pattern.

[0010] Many modern cameras, in particular, use HDR (High Dynamic Range) mode to increase their dynamic range. Temporal HDR is especially common, where multiple (e.g., three) image records with different exposure times are recorded rapidly in sequence, and portions of the images are internally combined into a single image. Thus, image areas of different brightness come from different portions of the image. For example, the darkest area (e.g., the black squares on a chessboard) comes from the first temporary portion, and the brightest area (the white squares on the chessboard) comes from the last temporary portion. If the camera moves relative to the target at the same time, a local displacement between the bright and dark areas of the target is produced in the image. The effect is particularly noticeable at the edges of light and dark, where all forms of blending can occur. This recording may not be suitable for precise calibration because the marker points can no longer be accurately determined. In the case of local HDR, large or sensitive pixels and small or insensitive pixels alternate on the sensor. This can lead to inaccuracies in determining the marker points because the pixels responsible for the bright and dark areas are located at different locations.

[0011] The point-symmetric regions and / or patterns presented herein can avoid or at least mitigate this undesirable effect as described above. By construction, these regions / or patterns can have a substantially uniform brightness distribution. All gray levels (each configurable) between the brightest and darkest gray values ​​can occur with substantially the same frequency in the region or pattern. True black and true white can be completely avoided, thus preventing nonlinear effects from the outset, particularly during pattern printing and during image recording and in-camera image preprocessing. When the region or pattern is positioned in front of the camera in an image-filling manner, it can be produced (e.g., printed) at a resolution at least as high as the camera's resolution. If the region or pattern is far from the camera, i.e., no longer in an image-filling manner, the resolution of the region and / or pattern is sufficient. Therefore, the point-symmetric regions or patterns can be smoothed by the imaging blur present in the optical path (through the lens, the image sensor, and, depending on the camera version, through an optional anti-aliasing filter in front of the image sensor). During this smoothing process, the distribution of grayscale values ​​changes: the originally given uniform distribution can tend to become a normal distribution with its center of gravity located in the middle grayscale region. This allows the camera to operate within its medium sensitivity range, for which the camera has been optimized, and artifacts can be avoided in the camera's dark extrema regions and especially in the camera's bright extrema regions.

[0012] A method for providing calibration data to calibrate a camera is proposed, wherein the method comprises the following steps:

[0013] Image data provided by the camera is read from the interface to the camera, wherein the image data represents a camera image of at least one predefined even and / or odd point symmetrical region in the environment of the camera;

[0014] The image data and determination rules are used to determine at least one center of symmetry for at least one even- and / or odd-numbered point symmetric region;

[0015] The position of the at least one center of symmetry in the camera image is compared with the predefined position of at least one reference center of symmetry in the reference image relative to the reference coordinate system to determine the positional deviation between the center of symmetry and the reference center of symmetry; and

[0016] The positional deviation is used to obtain displacement information of at least one subset of pixels in the camera image relative to corresponding pixels in the reference image, wherein the displacement information is used to provide the calibration data.

[0017] This method can be implemented, for example, in software or hardware, or a hybrid of software and hardware, such as in a control device or apparatus. At least one predefined point-symmetric region can be manufactured by performing a variation of the manufacturing method described below. The determination rules can be similar to or correspond to the process disclosed in DE 10 2020 202 160, which was subsequently published by the applicant. A reference image can represent at least one predefined point-symmetric region. The reference image can also be replaced by reference data that at least partially corresponds to or is equivalent to information obtainable from the reference image. Using reference data may be advantageous, particularly in the sense of reducing workload when information that can be extracted from the reference image has already been given in a more usable form in the type of reference data. The reference data can represent the reference image in a compressed form or representation, such as as a descriptor image, a signature image, and / or a list of all coordinates and types with existing centers of symmetry. The determination step can be performed using optical flow, particularly dense optical flow. Displacement information can represent displacement vectors or absolute coordinates.

[0018] According to one embodiment, the determination rule used in the determination step can be configured such that signatures are generated for multiple pixels of at least one segment of the camera image to obtain multiple signatures. In this case, descriptors with multiple different filters can be used to generate each signature. Each filter can have at least one symmetry type. Each signature can have a symbol for each filter of the descriptor. The determination rule can also be configured such that at least one mirror signature for at least one symmetry type of the filter is obtained for the signature. The determination rule can also be configured such that at least one additional pixel with a signature corresponding to at least one mirror signature is present in a search area in the environment surrounding the pixel with the signature, to obtain the pixel coordinates of at least one symmetry signature pair from the pixel and the additional pixel when the at least one additional pixel is present. In addition, the determination rule can be configured such that the pixel coordinates of the at least one symmetry signature pair are evaluated to identify the at least one symmetry center. The descriptor can describe the image content in the local environment surrounding the pixel or reference pixel in a compact form. The signature can, for example, represent the value of the descriptor describing the pixel in binary. Therefore, multiple calculated signature images can be used to determine the at least one mirrored signature; for example, one signature image with a normal filter, one signature image with an even-numbered point mirror filter, and one signature image with an odd-numbered point mirror filter. Additionally or alternatively, at least one reflector can be applied to the symbols of one of the signatures to obtain at least one mirrored signature. In this case, each reflector can have rules for filtering based on the symmetry type and descriptor to modify the symbols. Here, the search area can depend on at least one of the applied reflectors. Such an implementation provides the advantage of enabling efficient and accurate detection of symmetry features in image data. In this case, symmetry detection in an image can be achieved with minimal effort.

[0019] Here, in the determination step, for each determined center of symmetry, using the pixel coordinates of each symmetry signature pair that helps to correctly identify the center of symmetry, a transformation rule is determined for transforming the pixel coordinates of the center of symmetry and / or at least one predefined even- and / or odd-numbered point symmetry region. This transformation rule can be applied to the pixel coordinates of the center of symmetry and / or at least one predefined even- and / or odd-numbered point symmetry region to correct the distorted viewpoint of the camera image. The advantage of this implementation is that it allows for reliable and accurate reconstruction of the correct mesh or correct topology for multiple point symmetry regions.

[0020] The symmetry type of the at least one symmetry center can also be determined in the determination step. The symmetry type can represent even-numbered point symmetry and additionally or alternatively, odd-numbered point symmetry. Additionally or alternatively, in this case, in the comparison step, the symmetry type of the at least one symmetry center in the camera image can be compared with a predefined symmetry type of the at least one reference symmetry center in a reference image to check for consistency between the at least one symmetry center and the at least one reference symmetry center. Odd-numbered point symmetry can be generated by mirroring points with inverted grayscale or color values. By using and identifying two different point symmetries, the information content of the point symmetry region and pattern can be increased.

[0021] In this case, the image data read in the read-in step can represent a camera image of at least one pattern consisting of multiple predefined even- and / or odd-numbered symmetrical regions. Here, in the determination step, the geometric arrangement of the symmetry centers of the at least one pattern can be determined, a geometric sequence of the symmetry types of the symmetry centers can be determined, and additionally or alternatively, the pattern can be determined from multiple predefined patterns using the sequence. The arrangement and / or the sequence can represent an identification code for the pattern. This implementation offers the advantage of increased reliability in identifying symmetry centers and the ability to obtain further information by recognizing specific patterns. Reliable identification of symmetry centers can also be achieved for different distances between the camera and the pattern.

[0022] In this scenario, during the determination step, the arrangement of the symmetry centers of the at least one pattern, along with, additionally or alternatively, a sequence of symmetry types of the symmetry centers, is used to determine implicit additional information of the at least one pattern or readout rules for reading explicit additional information from the camera image. The arrangement and, additionally or alternatively, the sequence can represent the additional information in coded form. The additional information may be related to camera calibration. This implementation provides the advantage of conveying additional information through the topology of at least one pattern.

[0023] Here, in the comparison step, the reference image can also be selected from multiple stored reference images based on the determined arrangement, the determined sequence, and the additionally or alternatively determined pattern, or the reference image can be generated using stored generation rules. In this way, the correct reference image can be reliably identified. Optionally, when there is a link between the identified pattern and the generation rules, the memory requirements for the reference image can also be minimized, since only the generation rules need to be stored.

[0024] Furthermore, the determination steps and additional or alternative comparison steps can be performed jointly for all centers of symmetry, independent of their symmetry type, or the determination steps and additional or alternative comparison steps can be performed individually for centers of the same symmetry type, depending on their symmetry type. Therefore, joint execution allows for accurate and reliable identification of centers of symmetry with low memory and time requirements. Optionally, and particularly in particular, individual execution can minimize confusion with randomly occurring patterns in the image.

[0025] A method for calibrating a camera is also proposed, which includes the following steps:

[0026] Evaluate the calibration data provided according to the embodiments of the above method to generate a control signal dependent on the calibration data; and

[0027] The control signal is output to the interface of the camera or the camera's calibration device to calibrate the camera.

[0028] This method can be implemented, for example, in software or hardware, or a hybrid of software and hardware, for example, in a control device or apparatus. Here, the embodiments described above for the provided method can be advantageously combined to perform the calibration method.

[0029] Furthermore, a method is proposed for manufacturing at least one predefined even- and / or odd-point symmetric region for use in embodiments of the above method, wherein the method comprises the following steps:

[0030] Generate design data, which represents a graphical representation of the at least one predefined symmetrical region of even and / or odd points; and

[0031] The design data is used to generate the at least one predefined even and / or odd point symmetrical region on, at, or within the display medium to create the at least one predefined even and / or odd point symmetrical region.

[0032] This method can be implemented, for example, in software or hardware, or a hybrid of software and hardware, such as in a control device or apparatus. By performing this manufacturing method, at least one predefined even- and / or odd-numbered point symmetrical region can be manufactured, which can be used within the scope of the embodiments of the above-described method.

[0033] According to one embodiment, design data can be generated in the generation step, said design data being a graphical representation of at least one predefined even- and / or odd-numbered point symmetrical region represented by a circle, ellipse, square, rectangle, pentagon, hexagon, polygon, or annulus. In this case, the at least one predefined even- and / or odd-numbered point symmetrical region can have a regular or quasi-random content pattern. Additionally or alternatively, a first half of the at least one predefined even- and / or odd-numbered point symmetrical region can be arbitrarily predefined, and a second half can be constructed by dot mirroring and optionally additionally inverting grayscale values ​​and additionally or alternatively inverting color values. Additionally or alternatively, in the generation step, the at least one predefined even- and / or odd-numbered point symmetrical region can be generated by additive manufacturing processes, separation, coating, forming, initial forming, or optical display. Additionally or alternatively, the display medium can be glass, stone, ceramic, plastic, rubber, metal, concrete, plaster, paper, cardboard, food, or an optical display device. Therefore, at least one predefined even and / or odd point symmetric region can be manufactured in a suitable manner, depending on the specific purpose or application and the general boundary conditions therein.

[0034] Design data representing a graphical representation of at least one pattern composed of multiple predefined even- and / or odd-point symmetrical regions can also be generated in the generation step. In this case, at least a subset of the even- and / or odd-point symmetrical regions may be aligned on a regular or irregular grid, directly adjacent to each other, and additionally or alternatively separated from at least one adjacent even- and / or odd-point symmetrical region by gaps. They may be identical or different from each other in terms of their size and / or their content patterns, and additionally or alternatively arranged in a common plane or different planes. Additionally or alternatively, in the generation step, design data representing a graphical representation of at least one pattern with layered symmetry can be generated. In this way, different patterns with specific information content and additionally or alternatively patterns with layered symmetry can be generated for different distances from the pattern.

[0035] Even when corresponding markings are known to exist, it is particularly difficult for humans to perceive the symmetry hidden within a pattern. This, for example, makes it possible to conceal such markings. This may be meaningful or desirable for various reasons, such as aesthetic reasons, because technical markings should not or do not wish to be seen, for example, because attention should not be diminished by markings that are unimportant to humans, or because the markings should be kept confidential. Aesthetic reasons play an important role, especially in the field of design. For example, in the interior space of a vehicle, on the exterior of a vehicle, on an aesthetically pleasing object, or in the field of internal or architectural architecture, conspicuous technical markings are not or are difficult to accept. However, if technical markings are to be concealed, for example, in fabric patterns, or in plastic or ceramic reliefs, or in holograms, or on printed surfaces, as may be possible according to the implementation, the technical markings can be both aesthetically pleasing and useful, for example, providing one or more reference points for a camera, so as to determine, for example, the relative camera pose. Depending on the application, the concealed aspects may also be irrelevant or have little or no relevance. Thus, the robustness of the technology still applies to the use of such designed patterns. In particular, patterns with random or pseudo-random characters can provide a variety of possibilities for finding as clearly defined pairs of symmetrical points as possible. According to the implementation, this can be utilized, for example, particularly to improve the signal-to-noise ratio of the response measured at the center of symmetry, and thus to improve robustness in the sense of error-free detection and precise localization of the center of symmetry. The pattern may, in particular, comprise one or more point-symmetric regions having odd or even point symmetry. These regions may be designed, for example, as circles, hexagons, squares, ellipses, polygons, or other shapes. The point-symmetric regions may be of the same type or of different shapes and sizes. The point-symmetric regions may be connected to each other without gaps or spaced apart.

[0036] The proposed solution also creates a device configured to perform, manipulate, or implement variations of the methods presented herein within a corresponding apparatus. The task on which this invention is based can also be solved quickly and efficiently through this embodiment of the invention in the form of a device.

[0037] To this end, the device may have at least one computing unit for processing signals or data, at least one storage unit for storing signals or data, at least one interface to a sensor or actuator, at least one communication interface for reading sensor signals from the sensor or for outputting data or control signals to the actuator, and / or for reading or outputting data embedded in a communication protocol. The computing unit may be, for example, a signal processor, a microcontroller, etc., and the storage unit may be flash memory, EEPROM, or magnetic storage. The communication interface may be configured to wirelessly and / or wiredly read or output data, wherein a communication interface capable of reading or outputting wired data may, for example, electrically or optically read the data from a corresponding data transmission line or electrically or optically output the data to a corresponding data transmission line.

[0038] In the current context, a device can be understood as an electrical apparatus that processes sensor signals and outputs control signals and / or data signals based on those signals. The device may have an interface that can be configured as hardware and / or software. When configured as hardware, the interface may, for example, be part of a so-called system ASIC that contains various functions of the device. However, the interface may also be a separate integrated circuit or at least partially composed of discrete components. When configured as software, the interface may be a software module that exists, for example, on a microcontroller along with other software modules.

[0039] A computer program product or computer program having program code is also advantageous. This program code can be stored on a machine-readable carrier or storage medium (such as semiconductor memory, hard disk storage, or optical memory) and is used to execute, implement, and / or manipulate the steps of the method according to one of the above embodiments, particularly when the program product or program is running on a computer or device. Here, the method can be implemented as a hardware accelerator on a SoC or ASIC. Attached Figure Description

[0040] Embodiments of the proposed scheme are shown in the accompanying drawings and explained in more detail in the following description.

[0041] Figure 1 An embodiment of the device for provision, an embodiment of the device for calibration, and a schematic diagram of the camera are shown;

[0042] Figure 2 A schematic diagram of an embodiment of the equipment used for manufacturing is shown;

[0043] Figure 3 A flowchart illustrating an embodiment of the provided method is shown;

[0044] Figure 4A flowchart illustrating an embodiment of the calibration method is shown.

[0045] Figure 5 A flowchart illustrating an embodiment of the method for manufacturing is shown;

[0046] Figure 6 A schematic diagram of a display medium having a pattern composed of predefined point symmetrical regions is shown according to an embodiment;

[0047] Figure 7 A schematic diagram of a display medium having a pattern composed of predefined point symmetrical regions is shown according to an embodiment;

[0048] Figure 8 It shows that it has a source Figure 7 A schematic diagram of a display medium for a pattern, highlighting a pattern or a predefined point symmetrical area;

[0049] Figure 9 A schematic diagram of a predefined point-symmetric region according to an embodiment is shown;

[0050] Figure 10 A schematic diagram of a pattern composed of predefined point-symmetric regions is shown according to an embodiment;

[0051] Figure 11 A schematic diagram illustrating the use of a lookup table according to an embodiment is shown;

[0052] Figure 12 A schematic diagram of a voting matrix according to an embodiment is shown;

[0053] Figure 13 A schematic diagram illustrating an exemplary pattern arranged in a cube-like form according to an embodiment, showing the correct identification of the grid;

[0054] Figure 14 Shown from an oblique perspective Figure 6 The first part of the illustration is a schematic diagram of the pattern shown.

[0055] Figure 15 Showing from Figure 14 The first part of the illustration shows a pattern in which a predefined point-symmetric region is highlighted;

[0056] Figure 16 An example is shown. Figure 15 A schematic diagram of the pattern after view correction;

[0057] Figure 17 A schematic diagram of an embodiment with a layered symmetrical pattern is shown;

[0058] Figure 18 A schematic diagram of an embodiment with a layered symmetrical pattern is shown;

[0059] Figure 19 A schematic diagram of an embodiment with a layered symmetrical pattern is shown;

[0060] Figure 20 A schematic diagram of a pattern according to an embodiment is shown;

[0061] Figure 21 A schematic juxtaposition of embodiments of a traditional chessboard pattern and a pattern composed of symmetrical regions of predefined points is shown;

[0062] Figure 22 This illustrates the combined effects of fuzziness and nonlinearity. Figure 21 The chessboard pattern and its schematic diagram; and

[0063] Figure 23 They are shown respectively Figure 21 and Figure 22 The chessboard pattern and its schematic diagram, which indicate the center of symmetry of the pattern. Detailed Implementation

[0064] In the following description of advantageous embodiments of the invention, the same or similar reference numerals are used for elements shown in different figures and having similar effects, wherein repeated descriptions of these elements are omitted.

[0065] Figure 1 An embodiment of the device 120 for provision, an embodiment of the device 140 for calibration, and a schematic diagram of the camera 102 are shown. Figure 1 In the illustrations, the providing device 120 or the providing device 120 and the calibration device 140 are shown separately or arranged externally to the camera 102. The providing device 120 and the calibration device 140 are connected to the camera 102 in a manner capable of data transmission. According to another embodiment, the providing device 120 and / or the calibration device 140 may also be part of the camera 102 and / or may be combined with each other.

[0066] Camera 102 is configured to record camera images of its environment. In the environment of camera 102, exemplarily only predefined even- and / or odd-numbered point symmetry regions 110 with a center of symmetry 112 are arranged. Camera 102 is also configured to provide or generate image data 105 representing a camera image, wherein the camera image also shows the predefined even- and / or odd-numbered point symmetry regions 110.

[0067] The providing device 120 is configured to provide calibration data 135 for calibrating the camera 102. For this purpose, the providing device 120 includes a reading device 124, a determining device 126, an execution device 130, and a seeking device 132. The reading device 124 is configured to read image data 105 from the input interface 122 of the providing device 120 to the camera 102. Furthermore, the reading device 124 is also configured to forward the image data 105 representing a camera image to the determining device 126.

[0068] The determining device 126 of the providing device 120 is configured to determine the center of symmetry 112 of at least one point-symmetric region 110 using image data 105 and determining rule 128. Determining rule 128 will be discussed in more detail below. It should be noted that determining rule 128 is similar to or corresponds to the process disclosed in DE 10 2020 202 160, which was later disclosed by the applicant. The determining device 126 is also configured to forward the determined center of symmetry 112 to the execution device 130.

[0069] The execution device 130 is configured to compare the position of at least one center of symmetry 112 in a camera image with the predefined position of at least one reference center of symmetry in a reference image 115 relative to a reference coordinate system to determine a positional deviation 131 between the center of symmetry 112 and the reference center of symmetry. The execution device 130 is also configured to read in or receive the reference image 115 or reference data 115 from a storage device 150. The storage device 150 may be implemented as part of or separate from the providing device 120. Furthermore, the execution device 130 is configured to forward the positional deviation 131 to the obtaining device 132.

[0070] The obtaining device 132 is configured to subsequently use the positional deviation 131 to obtain displacement information 133 of at least one subset of pixels in the camera image relative to corresponding pixels in the reference image 115. The providing device 120 is configured to use the displacement information 133 to provide calibration data 135. More precisely, the providing device 120 is configured to provide calibration data 135 to the calibration device 140 via the output interface 138 of the providing device 120.

[0071] Calibration device 140 is configured to calibrate or control the calibration of camera 102. For this purpose, calibration device 140 includes an evaluation device 144 and an output device 146. Calibration device 140 is configured to receive or read calibration data 135 from providing device 120 via input interface 142 of calibration device 140. Evaluation device 144 is configured to evaluate the calibration data 135 provided by providing device 120 to generate a control signal 145 dependent on the calibration data 135. Evaluation device 144 is also configured to forward the control signal 145 to output device 146. Output device 146 is configured to output the control signal 145 to output interface 148 of camera 102 and / or calibration device 104 of camera 102 to calibrate or control the calibration of camera 102.

[0072] Specifically, determination rule 128 is configured such that signatures are generated for multiple pixels of at least one segment of the camera image to obtain multiple signatures. In this case, each signature is generated using a descriptor with multiple different filters. Each filter has at least one symmetry type. Each signature has a symbol for each filter of the descriptor. Determination rule 128 can also be configured such that at least one reflector is applied to the symbol of one of the signatures to determine at least one mirror signature for at least one symmetry type of the filter for that signature. In this case, each reflector includes a symmetry type-specific and descriptor-dependent filter rule for modifying the symbol. Determination rule is also configured such that it is checked whether a pixel with a signature exists in a search area in the environment surrounding the pixel, depending on the applied at least one reflector, and the at least one additional pixel has a signature corresponding to at least one mirror signature, to obtain the pixel coordinates of at least one symmetry signature pair from the pixel and the additional pixel when the at least one additional pixel exists. In addition, determination rule is configured such that the pixel coordinates of the at least one symmetry signature pair are evaluated to identify the at least one symmetry center.

[0073] According to one embodiment, the determining device 126 is configured to generate, for each determined center of symmetry 112, a transformation rule for transforming the pixel coordinates of the center of symmetry 112 and / or the point symmetry region 110 using pixel coordinates that help correctly identify the center of symmetry 112. The transformation rule is applied to the pixel coordinates of the center of symmetry 112 and / or the point symmetry region 110 to correct for distorted viewpoints in the camera image. Furthermore, it is advantageous to determine the transformation rule based on multiple, particularly adjacent, point symmetry regions 110, as this is more robust, more accurate, and less affected by noise, especially if these point symmetry regions lie in a common plane. The application of the transformation is particularly advantageous when considering the arrangement of multiple centers of symmetry 112.

[0074] According to one embodiment, the determining device 126 is further configured to determine the symmetry type of at least one center of symmetry 112. This symmetry type represents even-numbered-point symmetry and additionally or alternatively, odd-numbered-point symmetry. Additionally or alternatively, in this case, the executing device 130 is configured to compare the symmetry type of at least one center of symmetry 112 in the camera image with a predefined symmetry type of at least one reference center of symmetry in the reference image 115 to check for consistency between at least one center of symmetry 112 and at least one reference center of symmetry.

[0075] Specifically, image data 105 in this case represents a camera image of at least one pattern composed of multiple predefined point symmetry regions 110. Here, determining device 126 is configured to determine the geometric arrangement of the symmetry centers 112 of at least one pattern, determine the geometric sequence of the symmetry types of the symmetry centers 112, and / or use said sequence to determine the correct pattern represented by image data 105 from multiple predefined patterns. This arrangement and / or the sequence may represent an identification code of the pattern. According to one embodiment, determining device 126 in this case is configured to use the arrangement of the symmetry centers 112 of at least one pattern and / or the sequence of the symmetry types of the symmetry centers 112 to determine implicit additional information of the at least one pattern or readout rules for reading explicit additional information in the camera image. This arrangement and / or the sequence represents the additional information in coded form. The additional information relates to the calibration of camera 102. Additionally or alternatively, the actuator 130 is configured in this case to select a reference image 115 from a plurality of stored reference images or to generate a reference image 115 using stored generation rules, based on a determined arrangement, a determined sequence, and / or a determined pattern.

[0076] Figure 2 A schematic diagram of an embodiment of a manufacturing apparatus 200 is shown. The manufacturing apparatus 200 is configured to manufacture at least one predefined even- and / or odd-numbered point symmetric region 110 for use in manufacturing processes. Figure 1 The provision of equipment or similar equipment and / or Figure 1 The calibration equipment or similar equipment is used. For this purpose, the manufacturing apparatus 200 includes a generating device 202 and a producing device 206. The generating device 202 is configured to generate design data 204. The design data 204 represents a graphical representation of at least one predefined even- and / or odd-numbered point symmetric region 110. The producing device 206 is configured to use the design data 204 to produce at least one predefined even- and / or odd-numbered point symmetric region 110 on, at, or within a display medium, to manufacture at least one predefined even- and / or odd-numbered point symmetric region 110.

[0077] According to one embodiment, the generating device 202 is configured to generate design data 204, which is a graphical representation of at least one predefined even- and / or odd-numbered point symmetrical region 110 represented by a circle, ellipse, square, rectangle, pentagon, hexagon, polygon, or annulus. The at least one predefined even- and / or odd-numbered point symmetrical region 110 has a regular or quasi-random content pattern, and / or a first half-face of any predefined even- and / or odd-numbered point symmetrical region 110 is pre-given, and a second half-face is constructed by dot mirroring and / or inverting grayscale and / or color values. Additionally or alternatively, the generating device 206 is configured to generate at least one predefined even- and / or odd-numbered point symmetrical region 110 by additive manufacturing processes, separation, coating, forming, initial forming, or optical display. Additionally or alternatively, in this case, the display medium has glass, stone, ceramic, plastic, rubber, metal, concrete, plaster, paper, cardboard, food, or an optical display device.

[0078] According to one embodiment, the generating device 202 is configured to generate design data 204 representing a graphical representation of at least one pattern composed of a plurality of predefined even- and / or odd-numbered point-symmetric regions 110, wherein at least one subset of the point-symmetric regions 110 are aligned on a regular or irregular grid, directly adjacent to each other and / or separated from at least one adjacent point-symmetric region 110 by gaps, are the same as or different from each other in terms of their size and / or their content patterns, and / or are arranged in a common plane or in different planes. Additionally or alternatively, the generating device 202 is configured to generate design data 204 representing a graphical representation of at least one pattern having layered symmetry.

[0079] Figure 3 A flowchart illustrating an embodiment of a method 300 for providing calibration data to calibrate a camera is shown. The method 300 for providing this data can be used in this case. Figure 1 The method 300 for providing the information includes a reading step 324, a determining step 326, an execution step 330, and a obtaining step 332.

[0080] In read-in step 324, image data provided by the camera is read from the interface to the camera. The image data represents a camera image of at least one predefined even- and / or odd-numbered point symmetry region in the camera environment. Then, in determination step 326, the image data and determination rules are used to determine at least one center of symmetry for the at least one point symmetry region. Subsequently, in execution step 330, the position of the at least one center of symmetry in the camera image is compared with the predefined position of at least one reference center of symmetry in a reference image relative to a reference coordinate system to determine the positional deviation between the center of symmetry and the reference center of symmetry. Subsequently, in obtaining step 332, the positional deviation is used to obtain displacement information of at least one subset of pixels in the camera image relative to corresponding pixels in the reference image. The obtained displacement information is used to provide calibration data.

[0081] According to one embodiment, the image data read in step 324 represents a camera image of at least one pattern composed of multiple predefined point symmetrical regions. Here, in step 326, the geometric arrangement of the symmetry centers of at least one pattern is determined, a geometric sequence of symmetry types of the symmetry centers is determined, and / or the pattern is determined from multiple predefined patterns using the sequence. This arrangement and / or the sequence represents an identification code for the pattern. Optionally, step 326 and / or step 330 are performed jointly for all symmetry centers regardless of their symmetry type, or individually for symmetry centers of the same symmetry type, depending on their symmetry type.

[0082] Figure 4 A flowchart illustrating an embodiment of a method 400 for calibrating a camera is shown. The calibration method 400 can use... Figure 1 The calibration is performed using calibration equipment or similar equipment. Furthermore, the calibration method 400 can be combined with... Figure 3 The method or similar method used for calibration is performed. Method 400 for calibration includes an evaluation step 444 and an output step 446.

[0083] In evaluation step 444, the evaluation is based on... Figure 3 The calibration data provided by the method or similar method is used to generate a control signal that depends on the calibration data. Subsequently, in output step 446, the control signal is output to an interface of the camera or the camera's calibration device to calibrate the camera.

[0084] Figure 5 A flowchart illustrating an embodiment of a method 500 for manufacturing is shown. The method 500 can be performed to manufacture at least one predefined point-symmetric region for... Figure 3 The method or similar method used and / or provided Figure 4The calibration method or similar method may be used. The manufacturing method 500 may also be combined with or used... Figure 2 The manufacturing method 500 is performed using equipment or similar equipment used for manufacturing. The manufacturing method 500 includes a generating step 502 and a producing step 506.

[0085] In generation step 502, design data representing a graphical representation of at least one predefined point symmetry region is generated. Subsequently, in generation step 506, the design data is used to generate at least one predefined point symmetry region on, at, or within the display medium to manufacture at least one predefined point symmetry region.

[0086] Figure 6 A schematic diagram of a display medium 600 according to an embodiment is shown, comprising a pattern 610 consisting of predefined point symmetrical regions 110A and 110B. In this case, each predefined point symmetrical region 110A and 110B corresponds to or is similar to... Figure 1 The pattern 610 consists of only 49 predefined point symmetry regions 110A and 110B, as exemplarily shown in Part A of the illustration, and only eight predefined point symmetry regions 110A and 110B, as exemplarily shown in Part B of the illustration. In this case, the first predefined point symmetry region 110A has an odd number of point symmetries as its symmetry type, while the second predefined point symmetry region 110B has an even number of point symmetries as its symmetry type. In this case, a noise-like image pattern having the corresponding pattern 610 is printed on each display medium 600.

[0087] Based on Figure 6 This illustrates the use of symmetry in machine vision according to an embodiment, where symmetry can be designed to be difficult or almost imperceptible to humans, but simultaneously robust to the embodiment, locally accurate, and detectable with minimal computational effort. In this case, point symmetries are more or less hidden in pattern 610, and the observer can hardly identify these point symmetries. Through... Figure 6Predefined dot-symmetric regions 110A and 110B are highlighted graphically, allowing a human observer to identify these regions within a noisy image pattern on the display medium 600. The first part of the illustration A includes 49 exemplary circular dot-symmetric regions 110A and 110B, with only 25 of the first regions 110A having an odd number of dot-symmetries and 24 of the second regions 110B having an even number of dot-symmetries. In the second part of the illustration B, the dot-symmetric regions 110A and 110B are chosen to be larger than those in the first part of the illustration A, with only five of them having an odd number of dot-symmetries and only three of them having an even number of dot-symmetries, thus making them particularly suitable for larger camera distances or lower image resolutions. Therefore, the circular dot-symmetric regions 110A and 110B are positioned on the display medium 600, which is designed as a plate, where, in the case of odd or negative dot-symmetry, a bright dot mirror will appear as a dark image, and vice versa, while in the case of even or positive dot-symmetry, this reversal does not occur. If multiple patterns 610 are required, these patterns can be designed to be distinguishable. This can be accomplished by arranging the symmetrical centers of regions 110A and 110B, such as Figure 6 As shown, the first part of the illustration A and the second part of the illustration B are easily distinguishable, or based on the sequence of negative or odd point symmetry and positive or even point symmetry of regions 110A and 110B within the corresponding pattern 610.

[0088] Figure 7 A schematic diagram of a display medium 600 according to an embodiment is shown, having a pattern 610 composed of predefined point-symmetric regions. The pattern 610 in this case corresponds to or resembles a pattern from... Figure 6 One of the patterns, in which pattern 610 is in Figure 7 It is shown in the illustration but not highlighted graphically. Figure 7 The example shown is only an illustration of the relationship between the two. Figure 6 The carrier medium in the middle is similar to ten display media 600.

[0089] Figure 8 It shows that it has a source Figure 7 A schematic diagram of the display medium 600 with pattern 610, wherein the pattern or predefined point symmetrical regions 110A and 110B are graphically highlighted. By way of example only, the pattern 610 with predefined point symmetrical regions 110A and 110B is arranged or graphically highlighted on ten display media 600 in this case.

[0090] therefore, Figure 7 and Figure 8Ten patterns 610 optimized for distinguishability are shown only as an example. Each pattern 610 has a separate arrangement of odd-point symmetric regions 110A and even-point symmetric regions 110B. The pattern 610 is thus encoded through this arrangement. The encoding is chosen to be mutually coordinated and / or optimized through training so that even if the ten patterns 610 are captured by a camera rotated, mirrored, or partially hidden, these ten patterns remain clearly identifiable and distinguishable. Figure 7 and Figure 8 In pattern 610, the point symmetry regions 110A and 110B at the four corners of each display medium 600 are intentionally designed to be slightly more prominent. This is unrelated to the function itself, but provides practical advantages when manually assembling display media 600 with pattern 610. Display media 600 with pattern 610 can be arranged arbitrarily within the scope of the manufacturing methods already described, such as serially in three dimensions or planarly, or as a surface. Within the scope of the supply methods already described and / or by means of the supply devices already described, the point symmetry center of pattern 610 can be found correctly and accurately. Pattern 610 can, for example, be printed on a solid plate of any size, which can optionally be placed in an arrangement partially orthogonal to each other. Even in the case of blurred imaging of pattern 610 by camera, the symmetry center can be detected sufficiently well to achieve the described function. Therefore, the detection of the point symmetry center is robust for blurred imaging. This expands the application to situations where shallow depth of field is used, such as in low-light scenes, or when the camera's focus or autofocus is incorrectly set or cannot achieve perfectly sharp imaging, such as in liquids, turbid or moving media, in the edge areas of the lens, or during relative movement between pattern 610 and the camera (motion blur, directional blur). Even though point symmetry occurs naturally and especially in artificially designed environments, the potential false detections based on them differ spatially from those based on the correct pattern 610, and therefore the two sets can be easily separated or distinguished from each other.

[0091] To demonstrate that the method described above is also applicable to moving, non-flat, and even elastic surfaces, one can... Figure 7 and Figure 8 The pattern 610 is printed on, for example, paper and assembled into a flexible box. The method described above applies without problem even to non-flat or elastic surfaces (e.g., those made of paper). This makes it possible to determine the movement of these surfaces. Unlike many materials, paper, while not permissible to be cut, exhibits point symmetry that is invariant to shearing, thus shearing poses no problem.

[0092] In particular, the orientation of the center of symmetry in the camera image can be precisely determined. However, extending this precise measurement to the entire surface of pattern 610 may also be of interest in various applications. That is, each point or pixel of pattern 610 indicates where that point or pixel is located in the camera image. This then allows, for example, determining the minimum deviation between the actually observed pattern 610 and the ideal pattern based on the ground truth. For example, it is of interest to determine the accurate shape of pattern 610 when it is applied to a non-smooth or non-rigid surface by printing and thereby creating, for example, variable folds or indentations in pattern 610. Patterns with random characteristics are particularly well-suited for finding corresponding points from a first image to a second image. Here, the first and second images can be recorded chronologically from different perspectives using the same camera or two cameras.

[0093] Specifically, we should now consider a scenario where the first image is a real image from a camera and the second image is an artificially generated (stored) image of a given pattern (also called a reference image), which is placed (e.g., scaled, rotated, affine mapped, projected) onto the second image based on a found center of symmetry, such that it is as close as possible to the real (first) image. For the reference image, processing steps required for the first image from the camera, such as image preprocessing, are skipped or omitted if necessary. Known methods, such as optical flow or parallax estimation, can then be applied to find, for example, the corresponding pixel in the reference image for each pixel in the camera image—or vice versa. This results in a two-step process: in the first step, the found center of symmetry, along with any necessary encoding, is used to register or coarsely align the real image with the known pattern. This then represents initialization so that in the second step, for example, an optical flow method is used to precisely determine the minimum deviation in the sense of local displacement between the registered real image and the pattern, and, if necessary, for each point or pixel of the image or pattern 610. The smaller the search area, the less computational effort is required for the second step. The computational workload here is typically very small—due to the good initialization from the first step. Since both steps require very little computation, high pixel throughput is achieved on common computer platforms, defined as the product of the frame repetition rate [images / second] and the image size [pixels / image]. If local inconsistencies are not found, it can usually be explained by the occlusion of an object by the line of sight toward pattern 610. From this, the shape or outline of the occluded object can be inferred.

[0094] A reference image should be provided for the two steps described above. This can be achieved by maintaining an associated reference image for all patterns 610 under discussion in a memory. The resulting memory overhead can be reduced by storing only the relevant parameters needed to recalculate or generate the reference image when required. For example, pattern 610 can be generated according to simple rules using a quasi-random number generator. The term "quasi" here means that the random number generator actually operates according to deterministic rules, so its results are reproducible, which is advantageous here. Here, the rules should be understood, for example, as what diameters the symmetrical regions 110A and 110B have, how mirroring should be performed, and how pattern 610 is composed of multiple patterns with different levels of detail in a weighted manner, such that the pattern is well detectable at short, medium, and long distances. Thus, it is sufficient to store only the initialization data (seed) of the quasi-random number generator and, if necessary, the selection of rules for constructing pattern 610. With the help of this formation rule, the reference pattern can be generated repeatedly and identically when needed (and then deleted again).

[0095] In summary, the two-step process can be represented as follows. In the first step, the centers of symmetry are found and their symbols are determined. Here, the symbols represent the case distinction between odd and even symmetry. By comparing the symbol sequences, it can be determined which of the multiple patterns is involved. The symbol sequence of pattern 610 can also be called a code. This code can be described in a compact way and requires a maximum of 64 bits for pattern 610 with, for example, an 8×8 center of symmetry. For comparison purposes, all existing or considered codes should be stored. From this set, a code that is as consistent as possible with the observation is searched. This result is usually explicit. Even if the camera can only capture a portion of pattern 610, for example, due to occlusion, such a search is still possible because, in this example with an 8×8 center of symmetry, the codes provide a very large number of up to 2 64 This presents several possibilities, but the number of completed patterns 610 will be much smaller, thus providing a high degree of redundancy. For each stored code, information needed to generate a reference image, such as parameter and rule selection, should also be stored. This reference image is generated for the second step, for example, as needed, i.e., only generated when required, and only temporarily if necessary.

[0096] Based on the center of symmetry location found in the camera image coordinates in the first step and its corresponding known location in the reference image, a transformation rule can be calculated to map these coordinates to each other as well as possible, for example, using projection or affine mapping, which is optimized in the sense of least squares. Through this transformation and appropriate filtering of the image data, the two images can be transformed (distorted) into a common coordinate system, such as the coordinate system of the camera image, the coordinate system of the reference image, or any third coordinate system. Then, a more accurate comparison is made between the two images that have thus been aligned, for example, using optical flow methods. For example, for each pixel of the first image (preferably considering its environment), the best corresponding pixel in the second image with the environment is searched. The relative displacement of the corresponding positions can be expressed as displacement information, in particular as absolute coordinates or displacement vectors. Such displacement vectors can be obtained with sub-pixel precision, so the correspondence is usually not on the pixel grid but between pixel grids. This information allows for highly accurate analysis of the entire surface of pattern 610 captured in a camera image, for example, to analyze the deformation or distortion of pattern 610 or its carrier / display medium 600 using an elastic pattern, or to analyze imaging deviations in the optical path in the case of a rigid pattern.

[0097] If the searched correspondence is not found in the expected area, partial occlusion of pattern 610 can be inferred. The cause of occlusion could be, for example, an object located on pattern 610, or a second pattern that partially occludes the first pattern. Valuable information, such as the mask or outline of an object, can also be obtained from this occlusion analysis.

[0098] Figure 9 A schematic diagram of predefined point-symmetric regions 110A and 110B according to an embodiment is shown. In this case, each predefined point-symmetric region 110A and 110B corresponds to or is similar to a predefined point-symmetric region from one of the above figures. A second point-symmetric or even-numbered point-symmetric region 110B, including its center of symmetry 112, is shown in first part of the illustration A, and a first point-symmetric or odd-numbered point-symmetric region 110A, including its center of symmetry 112, is shown in second part of the illustration B. In this case, the predefined point-symmetric regions 110A and 110B represent regions formed by gray levels.

[0099] The use of point symmetry offers the following advantages over other forms of symmetry: point symmetry is preserved when the pattern and / or at least one predefined point-symmetric region is rotated about the viewing axis; point symmetry is also preserved when the pattern and / or at least one predefined point-symmetric region is tilted, i.e., at a tilted viewpoint. Rotation and tilting of the pattern and / or at least one predefined point-symmetric region do not cause problems for the detection of odd-numbered and even-numbered point symmetry, as both are preserved in the process. Therefore, the methods or approaches already mentioned above are also applicable to tilted viewpoints of the pattern or at least one predefined point-symmetric region. In the case of even-numbered point symmetry, grayscale or color values ​​are preserved, for example, when the points are mirrored.

[0100] and Figure 9 In the first part of the illustration A, at the center of symmetry 112 points, symmetrically, for each gray value g, the same partner gray value g is found. PG =g. In Figure 9 The second part of the illustration, Figure B, shows odd-point symmetry, where each grayscale value is inverted: for example, white becomes black and vice versa, light gray becomes dark gray and vice versa. In the example where the grayscale value g is in the interval 0 ≤ g ≤ 1, from... Figure 9 In the diagram, half of region 110A shown at the top is derived from the original grayscale value g based on g PU =1-g forms the grayscale value g that is mirrored by the point in the simplest possible way. PU Nonlinearity can also be integrated into this inversion, for example, gamma correction, to compensate for other nonlinearities in image display and image recording. The formation of suitable odd- or even-numbered point symmetric patterns is correspondingly straightforward. For example, Figure 9 In the diagram, half of the corresponding area 110A or 110B shown at the top is arbitrarily set or randomly generated. From this, it is then concluded that... Figure 9 The lower half of the diagram is shown, and it is mirrored by points, where the gray values ​​of the odd-numbered points are reversed or the gray values ​​of the even-numbered points are not reversed.

[0101] This observation or generation can also be extended to colored patterns and / or predefined point-symmetric regions. In this case, when the number of points is odd-point symmetric, the RGB values ​​after point mirroring can be formed by inverting the individual original RGB values, which is the simplest possibility, i.e., r PU =1-r (red), g PU =1-g (g represents green here)), b PU =1 - b (blue). Thus, for example, dark violet is imaged as light green, and blue is imaged as orange. A color pattern can represent more information than a monochrome pattern, which can be advantageous. A prerequisite for using this advantage is that color information is also used to convert the raw image (i.e., a color image from a camera or other imaging sensor) into a descriptor.

[0102] The specific implementation of pattern 610 and / or at least one predefined point symmetrical region 110 or 110A and / or 110B should also be discussed below with reference to the above figures.

[0103] Regarding the arrangement of pattern 610 and / or at least one predefined point symmetrical region 110 or 110A and / or 110B, for example, Figure 6 As shown, the point-symmetric regions 110 or 110A and / or 110B can be, for example, circular, and these regions can be mostly arranged in a regular grid in pattern 610. For example, the faces between the circular regions 110 or 110A and / or 110B can remain unused. Alternatives exist: for example, regions 110 or 110A and / or 110B can be square and connected to each other without gaps, thus using an entire face; or the symmetric regions 110 or 110A and / or 110B can be regular hexagonal faces, also connected to each other without gaps, thus using an entire face.

[0104] In this association, Figure 10 A schematic diagram of a pattern 610, comprising predefined point symmetric regions 110A and 110B according to an embodiment, is shown. The predefined point symmetric regions 110A and 110B in this case correspond to or are similar to... Figure 1 , Figure 6 and / or Figure 8 The symmetric region of the predefined points in the [reference]. Figure 10 Regions 110A and 110B in pattern 610 are both circular and arranged on a hexagonal grid. In this case, the distance between grid points or centers of symmetry can correspond to the diameter of the circle. Thus, the unused surface 1010 between regions 110A and 110B in pattern 610 can be minimized.

[0105] Other arrangements and shapes, such as rectangles, polygons, etc., are also possible, and they can be combined with each other in shape and / or size. For example, alternations of pentagons and hexagons, like a regular soccer ball. Shapes can also be arranged in other ways, such as rotation, with asymmetrical regions where necessary. The center of symmetry can also be located outside the point-symmetric region itself. This is the case, for example, when a torus is used as a shape. It is also not necessary for all point-symmetric regions to lie in a common plane. Instead, they can lie on different faces arranged in space, and these faces are also allowed to be non-flat.

[0106] Pattern 610 and / or at least one predefined point symmetrical region 110 or 110A and / or 110B can be formed in a variety of ways. Only a few examples are described below. Random or quasi-random patterns, such as noise patterns. By introducing low spatial frequency components, these patterns are formed such that they are still perceived as noise patterns with sufficiently high contrast at moderate to large distances from the camera. So-called white noise, i.e., uncorrelated grayscale values, is not suitable for this. Aesthetically pleasing, and where necessary, regular patterns, such as floral patterns, tendril patterns (leaves, branches, flowers), decorative patterns, mosaics, mathematical patterns, traditional patterns, onion patterns, patterns composed of iconic symbols (hearts, etc.), imitations of random patterns in nature (e.g., farmland, woodland, lawns, pebble beaches, sand, bulk materials (gravel, salt, rice, seeds), marble, rubble, concrete, brick, slate, asphalt surfaces, starry skies, water surfaces, felt, hammered paint, rusted iron sheets, sheepskin, scattered particles, etc.), and scene photographs with any content. To generate point-symmetric regions and / or patterns from this pattern suitable for the purposes described herein, half of the corresponding face is arbitrarily pre-given, and a second half is constructed by point mirroring, with grayscale or color values ​​inverted if necessary. See also [link to relevant documentation]. Figure 9 As a simple example.

[0107] There are countless possibilities regarding the material, surface, and manufacture of pattern 610 and / or at least one predefined point symmetrical region 110 or 110A and / or 110B. The following list is not exhaustive: black and white printing, grayscale printing, or multicolor printing on various materials; printing on or behind glass or transparent films; printing on or behind frosted glass or translucent films; relief in stone, glass, plastic, or rubber; relief in fired materials such as pottery, terracotta, or ceramics; relief casting in metal, concrete, or plaster; embossing on plastic or paper / cardboard; etching on glass, metal, or ceramic surfaces; milling in wood, cardboard, metal, stone, etc.; fired surfaces in wood or paper; photographic exposure of paper or other materials; temporary or decaying or water-soluble patterns for short-term application on plant materials, ash, sand, wood, paper, fruit, eggshells, the skin of other foods, etc.; display as holograms; display on monitors or displays (which may vary over time if necessary); display on LCD films or other display films (which may vary over time if necessary), etc.

[0108] Regarding the possibilities of relief manufacturing, such as milling, embossing, stamping, etc., it should be noted that the area should be perceived by the camera as having odd and / or even point symmetry. Therefore, it may be necessary to consider, for example, later lighting (e.g., light incident obliquely on the relief) and nonlinearities and other disturbances in optical imaging during the design phase. Whether the 3D shape or relief itself has an even and / or odd point symmetry type is not important; rather, it is the image recorded by the camera that shows this symmetry. Here, the direction of light incident or illumination and the reflection of light on the surface are also relevant and should be considered together in the design. Regarding image recording and lighting, it should be noted that the recording technology should be designed to be suitable for capturing pattern 610 and / or at least one predefined point symmetry region 110 or 110A and / or 110B. In particular, when there is rapid relative movement between pattern 610 and / or (one or more) areas 110 or 110A and / or 110B and the camera, it is recommended to use appropriate lighting (e.g., flashlight, strobe light, or bright LED light) so that the exposure time and motion blur in the resulting image can be kept minimal. For various applications, it is meaningful to apply pattern 610 and / or (one or more) areas 110 or 110A and / or 110B to a transparent or translucent surface. This allows pattern 610 and / or (one or more) areas 110 or 110A and / or 110B to be illuminated from one side and observed from the other. With this solution, interfering reflections of the light source on the display medium can be effectively avoided. In principle, there is freedom to choose the front or back of the carrier or display medium for the arrangement of pattern 610 and / or (one or more) areas 110 or 110A and / or 110B, the light source, and the camera. When selected, the risk of pattern 610 and / or (one or more) areas 110 or 110A and / or 110B or the camera being contaminated or pattern 610 and / or (one or more) areas 110 or 110A and / or 110B being worn can also come into play: thus, for example, it makes sense to apply pattern 610 and / or (one or more) areas 110 or 110A and / or 110B and the camera to the back, because they are better protected there from, for example, dust or water, or because pattern 610 and / or (one or more) areas 110 or 110A and / or 110B are protected there from mechanical wear.

[0109] A method, also used in embodiments, is disclosed in DE 10 2020 202 160, which is published later, to reliably and with minimal computational effort find symmetrical regions or patterns in an image. In this case, the original image, i.e., a color or grayscale image from a camera or other imaging sensor, is converted into an image of descriptors, wherein the descriptors are formed based on the local environment of the original image. Here, a descriptor is another representative form of the local image content, prepared in a form that is easier to process. More simply here, it is understood in particular to include information about the environment of a point, not just about the point itself, a high degree of invariance to brightness or illumination and its variations, and low sensitivity to sensor noise. The descriptor image can have the same resolution as the original image, such that approximately one descriptor exists for each pixel of the original image. Alternatively or additionally, other resolutions are also possible.

[0110] A signature is formed from a corresponding descriptor, represented as a binary word in a computer unit, or from multiple adjacent descriptors. The signature describes the local environment of the pixels in the original image as characteristically as possible. The signature can also be the same as a descriptor or a portion thereof. The signature is used as an address to access a lookup table. Therefore, if the signature consists of N bits, it can access a table of size 2. N A lookup table (i.e., 2 to the power of N). Advantageously, the word length N of the signature should not be chosen too large, as the storage requirements of the table grow exponentially with N: for example, 8 ≤ N ≤ 32. The signature or descriptor is constructed such that the signature symmetry can be determined using simple operations, such as a bitwise XOR operation on a subset of bits. Example: S P =s^R P Where s is a signature of length N, R P It is a point-symmetric (P) and its corresponding reflector (R). The symbol ^ represents the bitwise XOR operation. Therefore, the signature S P The point-symmetric counterpart of the signature 's'. This relationship also applies to the opposite direction.

[0111] If the construction of the descriptor or signature is fixed, the reflector is therefore automatically set (and constant). By applying it to any signature, that signature can be converted into its symmetric counterpart. An algorithm exists that can find one or more symmetric signature pixels for a given signature at the current pixel within an optional finite search window. The center of symmetry is then located in the middle of the line connecting the positions of these two pixels. Voting weights are output there or as close as possible and collected in a voting map. In the voting map, the output voting weights accumulate at the locations of the searched centers of symmetry. These centers of symmetry can thus be found, for example, by traversing the voting map to find the accumulation points. This works for point symmetry, horizontal axis symmetry, vertical axis symmetry, and other symmetries where needed, such as mirror symmetry on other axes, and rotational symmetry. More precise localization with sub-pixel accuracy can be achieved if the local environment is also included in the observation when evaluating the voting matrix to determine the accumulation points and precisely locating the centers of symmetry.

[0112] DE 10 2020 202 160 Figure 15 An algorithm is presented that can find point-symmetric correspondences with the currently observed signature. However, this paper only considers even-numbered point symmetries.

[0113] According to an embodiment, the algorithm is extended to odd-point symmetry. Particularly advantageous here is that odd-point and even-point symmetries can be determined simultaneously in a single common traversal. This saves time, as only one traversal of the signature image is needed instead of two, and reduces latency. When only one (instead of two) traversal is required, the streaming mode processing can provide the results of the symmetry search with significantly lower latency. Here, processing begins as soon as the first pixel data from the camera arrives, and the processing steps are executed intensively and sequentially. This means that the signature is computed as soon as the necessary image data from the local environment of the current pixel is available. A symmetry search is immediately performed on the newly formed signature. Once some parts of the voting matrix are complete (as is the case when these parts are no longer and will no longer be part of the search area), these parts can be evaluated immediately, and the found symmetry (strong symmetry center) can be output immediately. This process results in very low latency, typically corresponding to only a small number of image rows, depending on the height of the search area. Low latency is crucial if a rapid response is required, such as within an adjustment loop where the actuator influences the relative pose between the symmetric object and the camera. Memory is also saved. The voting map can be used for both even-point and odd-point symmetry forms, where the two symmetry forms or types with different signs participate in the voting. For example, the voting weight is subtracted for odd-point symmetry and added for even-point symmetry. This will be explained in more detail below. Furthermore, energy can be saved by conserving memory. The aforementioned low-latency implementation also results in only a small amount of intermediate data needing to be stored compared to the entire image. This memory-efficient operation is particularly important for cost-critical embedded systems and also leads to savings in energy requirements.

[0114] Figure 11 A schematic diagram illustrating the use of lookup table 1150 according to an embodiment is shown. Lookup table 1150 can be generated by... Figure 1 The specific means by which the equipment or similar equipment is provided is used. In other words, Figure 11 An example of the algorithm process for finding symmetric correspondences at search points is as follows: Figure 1 The equipment or similar equipment used for provision and / or Figure 3 A snapshot related to the method or similar method being provided. Specifically, Figure 11 The illustration in the document is also similar to that in the later published DE 10 2020 202 160. Figure 15 , among which here Figure 11 It also includes extensions to include even-point symmetry and odd-point symmetry.

[0115] Lookup table 1150, also known as an entry table, is shown. A pixel grid 1100 is illustrated, in which signatures with exemplary values ​​2412 are generated for the currently observed or processed pixel. In other words, Figure 11 A snapshot is shown during the formation of links of pixels or pixel coordinates with the same signature s. For clarity, two of at most N possible chains are shown, and this is for signature S. PG =364 and for signature S PU =3731. In pixel grid 1100, for each pixel, a reference to the position of the last previously signed image with the same signature value is stored. This creates links to positions with the same signature. Therefore, the signature value itself does not need to be stored. For each signature value, the corresponding entry position in pixel grid 1100 is stored in a lookup table 1150 or an entry table with N table fields. Here, N corresponds to the number of possible signature values. The stored value can also be "invalid". The contents of lookup table 1150 or the entry table and the referenced images (linked images) change dynamically.

[0116] In pixel grid 1100, pixels are processed row by row, for example from... Figure 11 Starting from the top left, as indicated by the arrow, and having currently progressed to the pixel with signature s=2412. Links between pixel positions having the same signature s are stored only for the first image region 1101. For the second image region 1102 in the lower image portion, links and signatures are not yet known at the indicated time point, and for the third image region 1103 in the upper image portion, links are no longer needed, for example, due to the limitations of the search area, where the link memory for pixels in the third image region 1103 can be freed again.

[0117] For the newly formed signature s, by applying reflector R PG Form an even-numbered point mirror signature S PG =364. Index PG represents point-symmetric, even. Index PU, representing point-symmetric, odd, is also used below. This value is used as the address in lookup table 1150 to find the address assigned to the same signature value S. PG The entry is found in the link at pixel position 364. At the indicated time point, lookup table 1150 includes two elements: the entry pixel position corresponding to signature s and a reference to that position indicated by the curved arrow. For clarity, other possible contents of lookup table 1150 are not shown. Signature value S PGThe link =364 includes three pixel positions, only shown here as an example. Two of these are located within search region 1104, which can also have a different form than shown here, such as a rectangle or a circle. Here, when traversing unidirectionally along the link, starting from the bottom, two symmetrical corresponding candidates for the two points located within search region 1104 are found. The third correspondence as the first element of the link symmetrical to the even-numbered point is not of interest here, as it is located outside search region 1104 and therefore too far from the current pixel position. If the number of symmetrical center candidates 1112 is not too large, the voting weight of the corresponding symmetrical center position can be output for each symmetrical center candidate 1112. The symmetrical center candidates 1112 are located at the positions of signatures s and the corresponding even-numbered point mirror images of signatures S, respectively. PG The middle of the connecting axis between them. If there is more than one candidate symmetry center 1112, the voting weight can be reduced respectively, for example, the reciprocal of the number of candidate symmetry centers can be used as the corresponding voting weight. Thus, the unclear candidate symmetry center is weighted less than the clear candidate symmetry center.

[0118] We will now consider and use odd-point mirror signatures. Figure 11 In the snapshot shown, the newly formed signature s is processed by applying another reflector R. PU Form an odd-numbered point mirror signature S PU =3731. Similar to the process described above for mirror signatures of even-numbered points, the same steps are performed for mirror signatures of odd-numbered points. The entry for the corresponding link is found using the same lookup table 1150. Here, lookup table 1150 points to the link symmetrically represented for signature 3731 at odd-numbered points. The first two pixel positions along this link again lead to the formation of symmetry center candidates 1112, as they are arranged in search area 1104 and because the number of candidate symmetry center candidates 1112 is not too large. The last pixel position along this link is located in the third image area 1103. This area is no longer needed because it can no longer enter the search area 1104 that slides line by line here.

[0119] If the next reference within the link points to the third image region 1103, traversal along the link can be terminated. Of course, traversal also terminates when the end of the link is reached. In both cases, it makes sense to limit the number of symmetry center candidates 1112; that is, if there are too many competing symmetry center candidates 1112, all symmetry center candidates 1112 are discarded. Furthermore, it makes sense to terminate traversal along the link early if neither its end nor the third image region 1103 can be reached after a pre-given maximum number of steps along the link. In this case, all symmetry center candidates 1112 found up to there should also be discarded.

[0120] The memory used for links in the third image region 1103 can be freed up again, so only the link memory needs to be reserved for the size of the first image region 1101. Therefore, the link memory requirement is generally low and here depends essentially only on one dimension of the search region 1104 (here, the search region height) and one dimension of the signature image (here, the signature image width).

[0121] The candidate for symmetry center 1112 or the candidate for symmetry center may not always fall exactly at the pixel location; instead, there are three other possibilities. Therefore, there are a total of four possibilities:

[0122] 1. The point or center of symmetry candidate 1112 falls on the pixel position.

[0123] 2. The point or symmetry center candidate 1112 falls in the middle between two horizontally directly adjacent pixel positions.

[0124] 3. The point or symmetry center candidate 1112 falls in the middle between two vertically directly adjacent pixel positions.

[0125] 4. The point or symmetry center candidate 1112 falls in the middle between four directly adjacent pixel positions.

[0126] In cases 2 through 4 where the details are unclear, it is advantageous to distribute the voting weights to be output evenly across the participating pixel locations. The output voting weights are then input into the voting matrix and summed or accumulated therein.

[0127] Here, not only positive voting weights but also negative voting weights are used simultaneously. Specifically, even-numbered symmetry is equipped with a different sign (positive) than odd-numbered symmetry (negative here). This leads to a clear result: in image regions without symmetry, which in practice largely represent the majority, the positive and negative voting weight outputs are approximately balanced, thus roughly canceling each other out in the voting matrix. Therefore, on average, approximately zero is obtained in the voting matrix. Conversely, in either odd-numbered or even-numbered symmetric regions, strong extrema are obtained in the voting matrix, and in this embodiment, a negative minimum is obtained when symmetric at odd points, and a positive maximum is obtained when symmetric at even points.

[0128] According to the embodiment shown here, the same resources are used for both odd-point symmetry and even-point symmetry, namely lookup table 1150 or entry table, link graph, voting matrix, which in particular saves memory requirements, and both symmetry forms or types are observed in a single common traversal, which saves time and intermediate memory.

[0129] Figure 12 A schematic diagram of a voting matrix according to an embodiment is shown in Table 1200. Table 1200 relates to a voting matrix, as a means of... Figure 1 A 3D image of a camera image processed by a provided device or similar device, in which the camera recorded from... Figure 6 The second part of the illustration shows the pattern. In the voting matrix or chart 1200, three exemplary maximum values ​​1210B and five minimum values ​​1210A can be clearly identified, representing those from... Figure 6 The second part of the illustration shows the symmetrical regions of three even-numbered points and five odd-numbered points. Outside of these extreme values, the values ​​in the voting matrix are close to zero. Therefore, the extreme values ​​can be determined very easily, and the location of the center of symmetry in the camera image can be determined explicitly and precisely.

[0130] Figure 12 These extreme values ​​are shown to be very obvious and therefore can be obtained through Figure 1 The equipment or similar equipment used for provision and / or Figure 3 The method or similar method used to provide this information is simple and unambiguous in detecting the symmetry. Here, information about the type of symmetry (i.e., odd or even) is included in the symbols. If the local environment of the corresponding extrema is also considered when evaluating the voting matrix, the location of the symmetry center can be determined with high precision at sub-pixel accuracy. Corresponding methods for this purpose are known to those skilled in the art. If the pattern is constructed appropriately, odd-point symmetry and even-point symmetry will not compete with each other. Thus, image regions (if any) will either have odd-point symmetry or even-point symmetry. Even if odd-point symmetric regions and even-point symmetric regions are close to each other in the camera image, it can be ensured that their symmetry centers remain spatially separated or distinguishable. Then, by jointly processing negative and positive symmetry, advantages in terms of resources and speed are achieved.

[0131] According to an embodiment, the processing of odd-point symmetry and even-point symmetry can be set up separately. Separating them before inputting entries into the voting matrix makes sense: two unsigned voting matrices are then set up to replace the common signed voting matrix, where the voting weights for negative symmetry are input into the first voting matrix and the voting weights for positive symmetry are input into the second voting matrix. This presents a potentially interesting advantage: a pattern can also be constructed and considered by the detection algorithm, which simultaneously exhibits odd-point symmetry and even-point symmetry with their centers of symmetry locally coinciding. While this mixed form of symmetry is highly unusual, this unusualness guarantees that it is extremely unlikely to be confused with randomly occurring patterns in the image. The two voting matrices are then searched to find the maximum value present at the same location in both matrices. Another possible advantage of processing odd-point symmetry and even-point symmetry separately is that it is easier to parallelize, thereby enabling faster execution when necessary. Because by using two voting matrices, access conflicts during voting weight input can be avoided, saving waiting time.

[0132] Figure 13 A schematic diagram of pattern 610, which is arranged in a cubic form according to an embodiment, is shown in terms of the correct identification of grid 1311. Figure 13 The pattern 610 shown is, for example, from Figure 7 or Figure 8 The pattern, in which three patterns are arranged in a cubic shape. In pattern 610, the detected or identified centers of symmetry 112A and 112B of the corresponding predefined point symmetry regions of pattern 610 are shown, wherein the signs and values ​​of the associated extrema in the voting matrix can optionally also be known. In this case, the first center of symmetry 112A is assigned to the predefined point symmetry region with odd-numbered point symmetry, and the second center of symmetry 112b is assigned to the predefined point symmetry region with even-numbered point symmetry. A correct grid 1311 is drawn for one of the patterns 610, on which the predefined point symmetry regions and thus the centers of symmetry 112A and 112B are aligned. For the other two patterns 610, the correct grid is searched, wherein in Figure 13 In the diagram, an incorrect solution to the grid search is indicated by the first label 1313, and a correct solution to the grid search is indicated by the second label 1314.

[0133] Finding the correct associated grid is a task with inherent fuzziness. After detecting the odd / even coded centers of symmetry 112A and 112B, the next step is typically to group them and determine which pattern 610 this group is assigned to, since it is not always known beforehand which patterns 610 and how many patterns 610 are included in the image. Part of this task could be finding the grid 1311 on which the centers of symmetry 112A and 112B are arranged. Instead of the square grid 1311, other topologies for arranging the centers of symmetry 112A and 112B are also considered, such as a concentric annular arrangement, see, for example... Figure 6 The second part of the illustration. As a representative example, observe square grid 1311 below.

[0134] Based on only Figure 13 The task of determining the correct grid positions for all patterns 610 by identifying the symmetry centers 112A and 112B is, in some cases, an ambiguous problem. If in Figure 13Observing pattern 610, if the correct grid 1311 has already been drawn for this pattern, it is not difficult to indicate the correct grid 1311 (to the observer). However, for the other two patterns 610 captured by the camera from a significantly more oblique perspective, it is clear that the output may be ambiguous. There are several possible solutions regarding how the grid can be placed through the centers of symmetry 112A and 112B. Here, the solution that is initially most obvious when viewed locally, i.e., the solution with an approximately vertical axis, is not the correct solution, as can be seen based on the first mark 1313. Instead, the second mark 1314 is correctly located on the grid. This shows that a naive process, such as searching for the nearest neighbor of the corresponding center of symmetry, may lead to an incorrect solution when viewed from an oblique perspective. In practice, solutions with a very oblique perspective are excluded because it is no longer possible to find the centers of symmetry 112A and 112B.

[0135] Figure 14 Shown from a tilted perspective Figure 6 The first part of the illustration shows a schematic diagram of pattern 610. Figure 14 The image shows a display medium 600 having a pattern 610 consisting of predefined point symmetrical regions 110A and 110B. Figure 14 The second part of the diagram, Figure B, shows the method of using... Figure 1 The equipment or similar equipment used for provision and / or Figure 3 The center of symmetry 112A and 112B of pattern 610 are used to identify or detect the pattern using the provided method or similar method. The center of symmetry 112A and 112B have been detected and at least their positions are available.

[0136] Figure 15 Showing from Figure 14 The first part of the illustration depicts pattern 610, in which predefined point symmetrical regions 110B are highlighted. Here, the predefined even-numbered point symmetrical regions 110B are highlighted graphically only as an example to illustrate the distortion of pattern 610 or regions 110A and 110B due to a tilted perspective. Here, the exemplary circular predefined point symmetrical regions 110A and 110B are distorted into ellipses by the tilted perspective.

[0137] The following is a special reference Figure 14 and Figure 15 Furthermore, the reconstruction of the correct mesh or topology of pattern 610 is discussed in general with reference to the above-mentioned figures.

[0138] From an oblique perspective, each circular region 110A and 110B from which the votes of the corresponding centers of symmetry 112A and 112B originate becomes an ellipse. This is achieved by backtracking on the corresponding centers of symmetry 112A and 112B (e.g., ...). Figure 15The voting, which contributes to the votes for the even-numbered point symmetry center 112B highlighted in the diagram, allows us to infer the shape and orientation of the corresponding ellipse. The direction and ratio of the ellipse's principal axis reveal how the ellipse can be stretched or straightened to transform it back into a circle. Observe the exemplary highlighted predefined even-numbered point symmetry region 110B in pattern 610, which contributes to the highlighted point symmetry center 112B. Depending on the design or construction, this region 110B is circular or approximately circular, such as hexagonal. From an oblique perspective, this circle becomes an ellipse. When voting is conducted to identify the symmetry center 112B, pairs of symmetric points help to form extrema in the voting matrix located within this ellipse.

[0139] According to one embodiment, the origin of point pairs in the camera image that lead to the formation of sufficiently strong extrema is traced. Further processing steps are performed for this purpose. First, it is assumed that voting has been conducted and a sufficiently strong center of symmetry has been found. Therefore, the starting point is as follows: Figure 14 The second part of the diagram illustrates the situation shown in B. The voting process is then iterated again in a modified form. However, the existing voting matrix is ​​not reformulated here. Instead, for each pair of symmetric points that contributes to the voting matrix, it is checked whether that contribution contributes to one of the found centers of symmetry 112A, 112B and therefore has already been contributed to in the first iteration. If so, the two positions of that pair are stored or immediately further calculated. Advantageously, the indices of the centers of symmetry 112A, 112B contributed by the symmetric point pair are also stored or used here. In this way, all contributions to the successful centers of symmetry can be determined afterward and (intermediately) stored or further used.

[0140] The further processing steps do not need to wait until the first processing step, namely the formation of the voting matrix and the determination of the centers of symmetry, is completed. Instead, they can begin earlier and utilize the intermediate results already achieved in the first processing step, namely the found centers of symmetry 112A and 112B. Then, from the information formed in this way, all image locations contributing to each found center of symmetry 112A and 112B can be read out. These locations are substantially, or except for a few outliers, located within the ellipse, such as... Figure 15 The example shown is with respect to the center of symmetry 112B.

[0141] Methods for determining the parameters of the ellipse are known to those skilled in the art. For example, a principal axis transformation can be formed over the set of all points contributing to the centers of symmetry 112A, 112B to determine the orientation of the principal axes and the two diameters of the ellipse. This can be achieved even without the need for intermediate storage of contributing image locations: instead, these image locations can be further processed immediately upon being known. Alternatively, an elliptical envelope can be determined around the set of points, using this elliptical envelope to enclose as closely as possible the largest possible portion of the set of points (excluding possible outliers).

[0142] Alternatively, an index image, equivalent to an index matrix, can be created instead of a list of points. The index image serves the same purpose—forming the parameters of all ellipses—but stores information in a different form. Ideally, the index image has the same dimensions as the signature image and is set to store indices, specifically those assigned to the found centers of symmetry 112A and 112B. Special index values, such as 0, are set to indicate that no entry exists yet. If a pair of symmetric points or signatures contributing to the i-th index is found during further processing steps, index i is entered at the two associated positions of the corresponding signature. Thus, at the end of the traversal, the following index image is obtained, in which all indices assigned to the centers of symmetry 112A and 112B appear multiple times, forming elliptical regions: thus, apart from a few outliers, each elliptical region contains only entries with uniform indices, and index 0 at unused positions. The index image can then be easily evaluated to determine the parameters of the individual ellipses. Furthermore, it is not necessary to store the entire index image. Once the data in a segment of the indexed image stops changing, that segment can be evaluated, and the memory can be freed up again. This results in lower latency, allowing intermediate results to be provided earlier.

[0143] The known elliptic parameters can then be used to correct the two-dimensional arrangement of the detected centers of symmetry (see...). Figure 14 This allows these centers of symmetry to subsequently lie on the grid of pattern 610, which here is at least approximately square, for example only.

[0144] Figure 16 The image after viewpoint correction is shown according to an embodiment. Figure 15 A schematic diagram of pattern 610. In other words, for illustrative purposes, Figure 16 It shows that in the Figure 15 Pattern 610 is orthogonal to or perpendicular to the direction of the found ellipse or the highlighted elliptical twisted region 110B, stretched by the ratio of the two principal axis lengths. Therefore, the correct grid 1311 can be found in a simple way. Thus, with... Figure 15In contrast, the ellipse is corrected by restoring the original circular shape of region 110B. Then, determining the grid 1311 where the centers of symmetry 112A and 112B are located, or determining the adjacency relationship between the centers of symmetry 112A and 112B without errors, is a straightforward matter. Figure 16 This is for illustrative purposes only. In practice, there is no need to distort the image. Since the information about the positions of the centers of symmetry 112A and 112B already exists in compressed form, it makes sense to use only this data for further processing and to transform its coordinates, where the transformation rule is formed by the determined ellipse parameters, and makes the ellipse into a circle.

[0145] When recording camera images at a telephoto focal length, a single global transformation is sufficient to determine grid 1311 for each segment. When recording camera images using a wide-angle lens (e.g., a fisheye lens), local transformations can be used at least in some areas. Therefore, the transformation rules described above can be applied globally and / or locally. In the global transformation, all projection centers are transformed using the same common transformation rule. This is meaningful and sufficient in many cases. The common transformation rule can be formed based on the common observation of all ellipses. If the centers of symmetry 112A and 112B lie on multiple planes in space, the ellipses can be grouped according to their parameters. Here, ellipses belonging to a plane have very similar parameters—especially when the plane is flat. A global transformation rule can then be determined and applied for each group. This process is applicable to telephoto focal lengths. Local transformations are meaningful when multiple circular regions are imaged as ellipses of different shapes or orientations by camera imaging. This is especially true for wide-angle cameras or high-torque lenses.

[0146] After the transformation is applied, the centers of symmetry belonging to the same face are at least approximately located on a common grid 1311. The next task is to assign the centers of symmetry 112A and 112B to grid locations. This can be done, for example, iteratively in small steps. For example, for centers of symmetry 112A and 112B, search for up to four nearest neighbors with approximately the same distance, for which see also [link to documentation]. Figure 13The search continues from these neighbors to more distant neighbors until all captured centers of symmetry 112A and 112B belonging to pattern 610 are assigned to a common grid 1311 or can be excluded from the common grid 1311. Therefore, if a center of symmetry is encountered during this search that does not match the grid 1311 just observed in terms of distance, these centers of symmetry are not recorded, as they may be outliers or centers of symmetry belonging to other faces. This iterative search can be repeated for other faces, such that eventually every center of symmetry 112A, 112B, except for outliers, is assigned to a face. For these faces, pattern 610 can then be identified preferably based on the binary codes associated with the centers of symmetry 112A and 112B, which are respectively contained in the symbols of the extrema.

[0147] Figure 17 A schematic diagram of an embodiment of a pattern 1710 with layered symmetry is shown. Pattern 1710 corresponds to or is similar to the pattern in the above figures. More specifically, by way of example only, pattern 1710 has a two-level layered structure consisting of four predefined point-symmetric regions 110A and 110B. According to the embodiment shown here, by way of example only, pattern 1710 has two predefined odd-point-symmetric regions 110A and two predefined even-point-symmetric regions 110B. In this case, pattern 1710 has an odd-point-symmetric structure overall. The even-point-symmetric regions 110B and the odd-point-symmetric regions 110A are located at the first layered level. The overall arrangement of the odd-point-symmetric pattern 110B is located at the second layered level. The center of symmetry 112 of the second layered level is represented by a quarter circle.

[0148] Figure 18 A schematic diagram of an embodiment of pattern 1810 with layered symmetry is shown. Figure 18 The pattern 1810 in the image is similar to that from Figure 17 The pattern. More precisely, Figure 18 Another example of a two-level hierarchical structure consisting of predefined point-symmetric regions 110B is shown. In the first hierarchical level, the predefined point-symmetric regions 110B are assumed to be point-symmetric themselves. In the second hierarchical level, there is an odd-numbered point symmetry at the level of pattern 1810, where the center of symmetry 112 is located at the center of the six-part hexagon shown for illustration. This odd-numbered symmetry is represented here as a reversal of the predefined point-symmetric regions 110B, for example, mirroring a dark symbol on a light background as a light symbol on a dark background.

[0149] Figure 19 A schematic diagram of an embodiment with a layered symmetrical pattern 610 is shown. In this case, pattern 610 is composed of... Figure 17 and Figure 18Patterns 1710 and 1810, or their inversions and / or dotted mirror forms, are constructed. For example only, pattern 610 has... Figure 17 The two patterns 1710 and Figure 18 The three-tiered structure consists of two patterns 1810. Patterns 1710 and 1810 are odd in number, and therefore are mirrored at the center of symmetry 112 of pattern 610, which is located at the center of the six-part hexagon shown for illustration. For example, Figure 19 The pattern 1710 shown in the lower right corner is the reverse of the pattern 1710 in the upper left corner. This layering principle can be continued at will, that is, a fourth level, a fifth level, and so on can be constructed.

[0150] The following is for reference. Figure 17 , Figure 18 and Figure 19 Further discussion is given of patterns with hierarchical symmetry. Symmetrical patterns 610, 1710, and 1810 can be constructed in multiple levels such that, for example, there are small self-symmetrical regions in the first hierarchical level, and common observation of them leads to symmetry in the next higher hierarchical level. Figure 17 and Figure 18 Both examples exemplify how to construct two-level layered patterns 1710 or 1810. Based on this, in Figure 19 A three-level layered pattern 610 was constructed. Therefore, in Figure 19The example includes three hierarchical levels. The third hierarchical level extends across the entire surface of pattern 610 (the area enclosed by dashed lines) and includes a center of symmetry 112. In the second hierarchical level, there are four patterns 1710 and 1810 (each enclosed by solid lines), each with a central center of symmetry (not explicitly shown here). According to the embodiment shown here, there are therefore 16 predefined point symmetric regions in the first hierarchical level, each with a center of symmetry. Here, the symmetry of the third hierarchical level is visible from a greater distance. During close proximity, the four symmetries of the second hierarchical level are also visible. At shorter distances, or if the capture resolution of pattern 610 is sufficient, the symmetry of the first hierarchical level also becomes visible. Therefore, visual servoing, such as visual control of a robot in the direction of pattern 610 or in any other direction, can be implemented over a large range of distances. If finer or lower hierarchical levels can already be captured, it is generally unnecessary to capture coarser or higher hierarchical levels. It is also unnecessary to capture all symmetries at the corresponding hierarchical level simultaneously; for example, at very short distances, it is simply impossible to capture the entire pattern 610 in a camera image. Clearly, even and odd symmetries can be chosen and combined with some degree of freedom. Additional information can also be included in this setting, specifically a bit allocated for the choice between odd and even symmetries, which can be transmitted to the capture system in this way. "Some degree of freedom" here means that the remaining form of the symmetry at the corresponding hierarchical level inevitably derives from the next higher hierarchical level. In other words, for example, in… Figure 18 In the middle, for the top row, the patterns "X" and "O" can be freely chosen. Then, the second row is inevitably derived, and here it is reversed because negative point symmetry was chosen at the next layer level.

[0151] Figure 20 A schematic diagram of pattern 610 according to an embodiment is shown. In the first part of the illustration A, Figure 20 Pattern 610 is shown as an example; pattern 610 is from... Figure 8 One of the patterns. Figure 20 The first part, illustration A, is an example of implicit additional information, 8.8 = 64 bits, derived here for illustrative purposes only, based on the symmetry type of predefined point-symmetric regions 110A and 110B of pattern 610, or the symbol of the associated point symmetry. In the second part, illustration B, Figure 20Pattern 610 is shown, which is constructed, by way of example, from four predefined point-symmetric regions 110A and 110B. For example, it is constructed from one predefined odd-point-symmetric region 110A and three predefined even-point-symmetric regions 110B on a square grid. Furthermore, in this case, a code matrix 2010 for explicit additional information is arranged in pattern 610. By way of example only, the implicit additional information from the first part of the illustration A is explicitly contained in the code matrix 2010. The predefined region 110A with odd-point symmetry here represents or marks the starting row of the 8×8 matrix, thereby explicitly setting the readout order.

[0152] The following is for reference. Figure 20 The delivery of implicit or explicit additional information will be discussed in more detail.

[0153] It may be useful or necessary to transmit additional information based on pattern 610 to a recipient, such as a computer, autonomous robot, etc. The additional information can be more or less extensive. Some illustrative examples of additional information include parking spots, charging stations, southwest-facing locations at 52°07'01.9"N9°53'57.4"E, left turns, speed limits of 20 km / h, lawnmower charging stations, etc. Various options exist for transmitting information using imaging sensors or cameras. In particular, a distinction can be made between implicitly included and explicitly included additional information; for this, see [reference needed]. Figure 20 The two examples illustrate this, one implicitly and one explicitly, providing 64 bits of additional information. Implicit additional information means that it is somehow contained within the pattern 610 itself, which is symmetrical, while explicit additional information is typically designed and captured separately from these patterns 610.

[0154] based on Figure 20 The first part of the diagram, Figure A, illustrates one possibility for transmitting implicit additional information: implicit additional information as binary code. Since the choice between odd-point symmetry and even-point symmetry is made for each symmetric region 110A and 110B when constructing pattern 610, additional binary information (corresponding to 1 bit) can be transmitted separately. If patterns with both odd and even-point symmetry are also allowed simultaneously, then the binary additional information becomes ternary additional information, i.e., three cases instead of two.

[0155] Another possibility for transmitting additional information is derived by using the non-uniform distance between the centers of symmetry of regions 110A and 110B, i.e., implicit additional information based on this arrangement. Then, with... Figure 20 The arrangements shown are different—in Figure 20The centers of symmetry are located on a square grid, and these centers of symmetry will be arranged irregularly, with additional information or a portion thereof encoded in this arrangement. Example: If the corresponding centers of symmetry are allowed to shift a fixed distance to the left / right and up / down, nine possible positions are obtained, whereby each center of symmetry can encode log2(9) = 3.17 bits of additional information. The tilt angle between the imaging sensor and pattern 610 does not pose a problem in any of the possibilities mentioned. For example, a subset of the centers of symmetry (e.g., the four outermost centers of symmetry in the corners) can be used to define the base grid for the coordinate system or rules. The deviation or binary / ternary code used for encoding is then related to this base grid.

[0156] The symmetric regions 110A and 110B used for implicit additional information should not be too small, so that they form sufficiently prominent extrema in the voting matrix. If a larger amount of additional information (especially static, location-based additional information) is to be transmitted to the receiver (e.g., a mobile robot), it is advantageous to explicitly encode this additional information.

[0157] exist Figure 20 The second part of the illustration, Figure B, shows how additional, particularly static, location-based information can be explicitly transmitted to a receiver (e.g., a mobile robot): for example, it can be agreed that additional information exists at specific coordinates in a coordinate system defined by a center of symmetry, encoded, for example, in binary (black / white) or other gradients (grayscale) or color. The process thus consists of two steps: in the first step, a field, such as code matrix 2010, is found based on odd and even symmetry, and the additional information is encoded in that field. In the second step, the field is read out, and thus the information contained therein. The tilted viewing angle between the imaging sensor and pattern 610 does not pose a problem here, because for the additional information to be read out, it is neither necessary for the fundamental vectors of the found coordinate system to be perpendicular to each other nor for these fundamental vectors to have the same length. Optionally, the image can also be corrected so that a Cartesian coordinate system subsequently exists. Optionally, a display can also be mounted in the field with pattern 610, which can transmit information that changes over time and / or transmits information through time changes, in addition to information that is static over time.

[0158] Through implicit error detection, high-resolution supplementary information can also be included in the pattern 610 itself. Therefore, there is another possibility that supplementary information (particularly static, position-based) is transmitted via the pattern 610 itself: this means that supplementary information is contained within the sequence of the black-and-white, color, or grayscale pattern 610 itself. Through the above classification, this supplementary information will be both implicit and explicit. Since the pattern 610, or at least some of its parts, is symmetrical, supplementary information is automatically and redundantly included, typically doubly included separately. This applies to both odd-point and even-point symmetries. This fact can be used for error correction or error detection. For example, if the pattern 610 is contaminated, for example, with bird droppings, errors resulting from this can be detected with high reliability, because the same error is unlikely to appear at the associated symmetrical location.

[0159] The following discussion will specifically reiterate the topic of... Figure 1 Equipment or similar equipment used for calibration and / or Figure 4 Background and embodiments of methods or similar methods used for calibration.

[0160] The following points should be noted regarding the background of camera calibration. When calibrating a camera, a distinction is made between intrinsic calibration and non-intrinsic calibration. Intrinsic calibration refers to capturing and describing the image on the image sensor through the lens. Here, parameters such as focal length and principal image point are determined, as well as the deviation of the optical image from an ideal model (e.g., a pinhole camera model). In non-intrinsic calibration, the external orientation of the camera is determined, i.e., its position and orientation, such as relative to a given coordinate system (e.g., a vehicle coordinate system or a second camera). For calibration, a specific calibration target is recorded multiple times with the camera, where a calibration pattern is applied, for example, to a moving plate or three-dimensional volume. Typically, the relative arrangement between the camera and the calibration target varies between these records. The images captured by the camera to be calibrated are usually automatically evaluated. Here, the orientation of intersections or corners (e.g., in a checkerboard pattern) or centers, etc., is automatically determined. These points are also referred to hereinafter as marker points. The calibration parameters sought are then determined based on multiple such records and the marker points determined therefrom, for example, by means of bundle compensation methods. Typically, the calibration target is captured in more than one image. It is then desirable to identify whether it is the same target. Furthermore, it is meaningful to also capture the orientation of the target to eliminate ambiguity. For example, a simple checkerboard pattern with 8×8 squares becomes ambiguous when rotated 180° around the surface normal. Orientation is particularly important because targets are not ideal in practice; for example, no plane is perfect, so deviations must be estimated along with the results if high accuracy is to be expected.

[0161] To enable automatic and explicit determination of targets and their orientation, they are typically encoded. For this purpose, for example, a smaller complementary-color circle (i.e., a black circle in a white square or a white circle in a black square) can be inserted within a portion of a square or circle. Binary information (1 bit) can be provided regarding its presence or absence. The gaps between the circles can also, for example, contain small circles for encoding. Other variations of this encoding exist, such as using QR codes, ArUco codes, proprietary codes, etc. The encoding can also be contained only within the edges, rather than distributed across the entire surface. Then, for example, a common checkerboard pattern can be surrounded by a code border. However, a drawback of such a target is that a portion of its surface needs to be used only for encoding and therefore cannot be used for calibration via markers. The bright surface of the corresponding pattern can often also be designed to be retroreflective to achieve high luminous efficiency on the calibration target using a light source close to the camera. For example, a random pattern composed of noise patterns of different spatial frequencies can also be used for calibration. Therefore, while this random pattern may appear at first glance similar to… Figure 6 The pattern 610 or a similar pattern, but without point symmetry.

[0162] Figure 21 A schematic juxtaposition of embodiments of a conventional chessboard pattern 2100 and a pattern 610 composed of predefined point symmetrical regions 110A and 110B is shown. Figure 21 In the first part of the illustration A, a chessboard pattern 2100 is shown, and in the second part of the illustration B, a chessboard pattern 2100 is shown. Figure 6 The first part of the illustration shows pattern 610 or a similar pattern, along with highlighted predefined point symmetry regions. The third part, illustration C, shows pattern 610 with highlighted predefined point symmetry regions 110A and 110B. By way of example only, pattern 610 includes 25 first predefined point symmetry regions 110A with odd-numbered symmetry and 24 second predefined point symmetry regions 110B with even-numbered symmetry.

[0163] Figure 22 This illustrates the combined effects of fuzziness and nonlinearity. Figure 21 A schematic diagram of chessboard patterns 2100 and 610. More precisely, in Figure 22 The diagram illustrates the effects of a combination of fuzziness and nonlinearity on the data from... Figure 21 The first part of the diagram is shown in Figure A, and the second part is shown in Figure B. In other words, Figure 22 This includes an explanation or simulation of effects occurring in optical systems: a combination of blurring and nonlinearity. These result in the "burn-out" or "lamp-eating" effect known from photography, where bright areas extend into the neighborhood. This makes the white checkerboard squares appear larger than the black checkerboard squares.

[0164] Figure 23 They are shown respectively Figure 21 and Figure 22 A schematic diagram of chessboard patterns 2100 and 610, having symmetry centers 112A and 112B of the identified pattern 610, similar to... Figure 14 The second part of the illustration. In other words, Figure 23 This demonstrates the locatability of marker points in the disturbed image. Although the intersections on the chessboard pattern 2100 in Part A of the first illustration can no longer be definitively determined, the centers of symmetry 112A and 112B of pattern 610 in Part B of the second illustration can be found in a simple and accurate manner.

[0165] In order to illustrate at least a portion of the described effect, in Figure 21 , Figure 22 , Figure 23 The traditional chessboard pattern 2100 and the planar pattern 610 with embedded point symmetry were compared. Figure 21 The third part of the diagram, C, reveals that these are symmetrical regions 110A and 110B of 7×7 circular dots. Therefore, the number of marked points corresponds to the number on the chessboard pattern 2100, since 8×8 squares produce 7×7 intersections. For Figure 22 Two practically relevant effects were simulated: fuzziness and nonlinearity. If we observe the magnified chessboard pattern 2100, we can see that the previously marked points (intersections) are blurred and therefore ambiguous because white squares "span across" into adjacent black squares, as is known from practice. Locally determining the marked points becomes a guessing game, especially when additional effects such as noise are added. While pattern 610 in... Figure 22 The center appears to have changed, but the centers of symmetry 112A and 112B can be found effortlessly, as shown below. Figure 23 As shown, despite the presence of interference, locating the marker points is not a problem, and all of these marker points are precisely located on the grid. Advantageously, the orientation of the corresponding marker points does not need to be determined locally; rather, each complete surface segment contributes to the location of the centers of symmetry 112A and 112B. Effects acting on this surface segment (such as blurring or nonlinearity) do not affect the orientation of the centers of symmetry 112A and 112B. In this respect, the proposed pattern 610 is more invariant in terms of marker point location relative to typical interferences during image recording compared to conventional black-and-white patterns with squares and circles. According to the embodiment, this thus enables camera calibration and dense calibration using the point-symmetric or planar point-symmetric pattern 610.

[0166] If a pattern with, for example, 7×7=49 marker points (as in some of the previous examples) is used for calibration, it is clear that conclusions about the optical imaging can only be obtained for the 49 lines of sight between the points on the target and their images on the sensor. Therefore, in practice, dozens or hundreds of images will be recorded for calibration, with the relative arrangement between the camera and the target repeatedly changed during this process. However, it is obvious that the coverage will still be sparse. Therefore, interpolation should be performed on the pixels that were not captured. This corresponds to the conventional process.

[0167] According to an embodiment, this can be improved and dense calibration can be determined, meaning that the interpolation required so far can be replaced by measurements for each pixel. To this end, centers of symmetry 112A, 112B are found in a given camera image. These centers of symmetry represent marker points. The centers of symmetry 112A, 112B are assigned to one or more patterns 610, depending on whether the one or more patterns 610 are located in the image. Multiple patterns 610 are often involved in calibration, and these patterns can form multiple faces of one or more targets. Only one pattern is representatively observed below. Pattern 610 is identified based on included encoding, for example, also referencing... Figure 20 The explanation provided is provided. References may also be applied in this case. Figures 13 to 16 The described process for localizing the viewpoint correction causes a circle that has been compressed into an ellipse to revert to a circle. At this point, for each found center of symmetry 112A, 112B, its location on the identified pattern 610 is known; that is, a point-to-point mapping exists. A reference pattern image is loaded from memory based on code. Alternatively, this reference pattern image can also be generated on demand, as the formation rules require significantly less storage space than the reference pattern image itself. In this way, for example, in calibration software or in a database, a large number of formation rules for different patterns can be stored without consuming a large amount of storage space.

[0168] Therefore, it is now possible to compare the pattern 610 captured by the camera to be calibrated with a reference pattern, and to compare it point-by-point or pixel-by-pixel, i.e., densely. This can be done in the coordinates of the camera image, the coordinates of the reference image, or another independent coordinate system. The known point-to-point mapping of the marker points makes it possible to register the two images to be compared. In other words, the mapping is known for a few (e.g., 49) points, but not for millions of other pixels. Interpolation can then be started, for example, using the known marker points as support points. This interpolation provides an approximate solution to the mapping of the observed pattern 610 to the reference pattern. This approximate solution is used as initialization for a subsequent search. This search can be performed using dense optical flow techniques known to those skilled in the art. The goal of this search is to be able to indicate which point, pixel, or subpixel of the reference pattern each point, pixel, or subpixel of the observed pattern 610 is mapped to—or vice versa if necessary. Good initialization is advantageous in this case because the remaining uncertainty and therefore the remaining search area are small. Correspondingly, the search workload is also small, and the calibration methods can be performed quickly and / or using small computer units.

[0169] The result is a dense mapping that, for each point of the observed pattern 610, illustrates a measurement, not just interpolation, mapping to a point in the reference pattern, or vice versa. This dense mapping is used for calibration calculations. Dense calibrations formed in this way are particularly accurate and require a minimal number of camera records, especially compared to conventional calibrations, because for each record, not only information for the marked points is obtained, but also information for all points of the corresponding pattern 610.

[0170] If an embodiment includes an "and / or" link between a first feature and a second feature, it should be understood that the embodiment has both the first feature and the second feature according to one implementation, and either only the first feature or only the second feature according to another implementation.

Claims

1. A method (300) for providing calibration data (135) to calibrate a camera (102), wherein the method (300) comprises the following steps: Image data (105) provided by the camera (102) is read (324) from the interface (122) to the camera (102), wherein the image data (105) represents a camera image of at least one predefined even and / or odd point symmetric region (110; 110A, 110B) in the environment of the camera (102), wherein for even point symmetric regions, the gray values ​​of the point reflections of the region are preserved respectively, and for odd point symmetric regions, the gray values ​​of the point reflections are reversed respectively; The image data (105) and determination rules (128) are used to determine (326) at least one center of symmetry (112; 112A, 112B) of the at least one even and / or odd point symmetric region (110; 110A, 110B). The position of the at least one center of symmetry (112; 112A, 112B) in the camera image is compared (330) with the predefined position of at least one reference center of symmetry in the reference image (115) relative to the reference coordinate system to determine the positional deviation (131) between the center of symmetry (112; 112A, 112B) and the reference center of symmetry; and The positional deviation (131) is used to obtain (332) displacement information (133) of at least one pixel subset of the camera image relative to the corresponding pixel of the reference image (115), wherein the displacement information (133) is used to provide the calibration data (135).

2. The method (300) according to claim 1, wherein the determination rule (128) used in the determination step (326) is constructed such that A signature(s) is generated for multiple pixels of at least one segment of the camera image to obtain multiple signature(s), wherein each signature(s) is generated using a descriptor having multiple different filters, wherein each filter has at least one symmetric type, and wherein each signature(s) has a symbol for each filter of the descriptor. For the signature(s), at least one mirror signature (S) of at least one symmetric type is obtained for the filter. PG S PU ), Check whether the pixel having the signature(s) exists in at least one other pixel in the search area (1104) of the environment surrounding the pixel, the at least one other pixel having a corresponding mirror signature(s). PG S PU The signature(s) is used to obtain the pixel coordinates of at least one symmetric signature pair from the pixel and the other pixel when at least one other pixel is present. And evaluate the pixel coordinates of the at least one symmetric signature pair to identify the at least one center of symmetry (112; 112A, 112B). And / or at least one reflector (R) PG R PU The symbol applied to one of the signatures(s) is used to obtain the at least one mirror signature(S). PG S PU ), where each reflector (R) PG R PU ) has rules for modifying the symbol that are specific to the symmetry type and depend on the descriptor, wherein the search region (1104) depends on the applied reflector (R PG R PU At least one reflector in ).

3. The method (300) according to claim 2, wherein in the determining step (326), for each determined center of symmetry (112; 112A, 112B), using the pixel coordinates of each symmetry signature pair that has contributed to correctly identifying the center of symmetry (112; 112A, 112B), a transformation rule is determined for transforming the pixel coordinates of the center of symmetry (112; 112A, 112B) and / or the at least one even and / or odd point symmetry region (110; 110A, 110B), wherein the transformation rule is applied to the pixel coordinates of the center of symmetry (112; 112A, 112B) and / or the at least one even and / or odd point symmetry region (110; 110A, 110B) to correct the distorted viewpoint of the camera image.

4. The method (300) according to any one of claims 1 to 3, wherein in the determining step (326), the symmetry type of the at least one symmetry center (112; 112A, 112B) is determined, wherein the symmetry type represents even-point symmetry and / or odd-point symmetry, and / or in the comparing step (330), the symmetry type of the at least one symmetry center (112; 112A, 112B) in the camera image is compared with a predefined symmetry type of at least one reference symmetry center in the reference image (115) to check the consistency between the at least one symmetry center (112; 112A, 112B) and the at least one reference symmetry center.

5. The method (300) according to claim 4, wherein the image data (105) read in the reading step (324) represents a camera image of at least one pattern (610; 1710, 1810) consisting of a plurality of predefined even and / or odd point symmetric regions (110; 110A, 110B), wherein the geometric arrangement of the symmetry centers (112; 112A, 112B) of the at least one pattern (610; 1710, 1810) is determined in the determining step (326), the geometric sequence of the symmetry type of the symmetry centers (112; 112A, 112B) is determined, and / or the pattern (610; 1710, 1810) is determined from a plurality of predefined patterns using the sequence, wherein the arrangement and / or the sequence represents an identifier of the pattern (610; 1710, 1810).

6. The method (300) according to claim 5, wherein in the determining step (326), the arrangement of the symmetry centers (112; 112A, 112B) of the at least one pattern (610; 1710, 1810) and / or the sequence of the symmetry types of the symmetry centers (112; 112A, 112B) are used to determine implicit additional information of the at least one pattern (610; 1710, 1810) or readout rules for reading explicit additional information in the camera image, wherein the arrangement and / or the sequence represent the additional information in an encoded form, wherein the additional information relates to calibrating the camera (102).

7. The method (300) according to any one of claims 5 to 6, wherein in the comparison step (330), the reference image (115) is selected from a plurality of stored reference images according to the determined arrangement, the determined sequence and / or the determined pattern (610; 1710, 1810) or the reference image (115) is generated using stored generation rules.

8. The method (300) according to any one of claims 5 to 6, wherein the determination step (326) and / or comparison step (330) are performed jointly for all centers of symmetry (112; 112A, 112B) independently of the symmetry type of the centers of symmetry (112; 112A, 112B), or the determination step (326) and / or comparison step (330) are performed individually for centers of symmetry (112; 112A, 112B) of the same symmetry type according to the symmetry type of the centers of symmetry (112; 112A, 112B).

9. A method (400) for calibrating a camera (102), wherein the method (400) comprises the following steps: The evaluation (444) of the calibration data (135) provided by the method (300) according to any one of claims 1 to 8 is used to generate a control signal (145) dependent on the calibration data (135); and The control signal (145) is output to the interface (148) of the camera (102) or the calibration device (104) of the camera (102) to calibrate the camera (102).

10. A method (500) for manufacturing at least one predefined even- and / or odd-point symmetric region (110; 110A, 110B) for use in the method (300) according to any one of claims 1 to 8, wherein the method (500) comprises the following steps: Generate (502) design data (204), the design data representing a graphical representation of the at least one predefined even and / or odd point symmetric region (110; 110A, 110B); and Using the design data (204), at the display medium (600) or in the display medium (600), generate (506) the at least one predefined even and / or odd point symmetric region (110; 110A, 110B) to manufacture the at least one predefined even and / or odd point symmetric region (110; 110A, 110B).

11. The method (500) according to claim 10, wherein design data (204) is generated in the generation step (502), the design data being a graphical representation of the at least one predefined even and / or odd point symmetric region (110; 110A, 110B) as a circle, ellipse, square, rectangle, pentagon, hexagon, polygon, or annulus, wherein the at least one predefined even and / or odd point symmetric region (110; 110A, 110B) has a regular or quasi-random content pattern, and / or wherein any of the at least one predefined even number is pre-given. The first half of the and / or odd-numbered point symmetrical region (110; 110A, 110B) is constructed, and the second half is constructed by point mirroring and / or inversion of grayscale values ​​and / or color values, and / or wherein the at least one predefined even-numbered and / or odd-numbered point symmetrical region (110; 110A, 110B) is generated in the generation step (506) by additive manufacturing process, separation, coating, forming, primary forming or optical display, and / or wherein the display medium (600) has glass, stone, ceramic, plastic, rubber, metal, concrete, plaster, paper, cardboard, food or optical display device.

12. The method (500) according to any one of claims 10 to 11, wherein design data (204) is generated in the generation step (502), the design data representing a graphical representation of at least one pattern (610; 1710, 1810) consisting of a plurality of predefined even and / or odd point symmetric regions (100; 110A, 110B), wherein at least one subset of the point symmetric regions (100; 110A, 110B) is aligned on a regular or irregular grid (1311), directly adjacent to each other and / or separated from at least one adjacent even and / or odd point symmetric region (110; 110A, 110B) by gap portions, and is the same or different from each other in terms of their size and / or their content patterns and / or arranged in a common plane or in different planes, wherein the design data (204) representing a graphical representation of at least one pattern (610; 1710, 1810) having layered symmetry is generated in the generation step (502).

13. An apparatus (120; 140; 200) configured to perform and / or manipulate the steps of the method (300; 400; 500) according to any one of claims 1 to 12 in corresponding units (124, 126, 130, 132; 144, 146; 202, 206).

14. A computer program product configured to perform and / or manipulate the steps of the method (300; 400; 500) according to any one of claims 1 to 12.

15. A machine-readable storage medium having a computer program product according to claim 14 stored thereon.

Images