Method and device for inspecting a reflective surface

Abstract

The invention relates to methods for inspecting a reflective surface (200) of an object arranged in a measurement chamber, with a device which has at least one 2-dimensional illumination pattern (201) for reflection on the reflective surface (200) and a plurality of cameras (203) for pixel-by-pixel recording of image information of the illumination pattern (201) reflected on the surface (200), wherein the position and orientation of the plurality of cameras (203) and of the at least one 2-dimensional illumination pattern (201) are known in a common coordinate system and the illumination pattern represents a temporally varying illumination pattern, wherein object image information is generated by applying a measuring sequence in which the cameras capture sections of the measurement chamber. The object image information is processed by means of the evaluation device, in such a way that output data is generated therefrom with regard to an object-related inspection task taking into account determined segmentation information, wherein the inspection task comprises, for example, the generation of defect information on the object and / or the calculation of surface normals for a plurality of points in at least one section of the reflective surface of the object and / or the calculation of the curvature for a plurality of points in at least one section of the reflective surface of the object, as output data. The invention further relates to a corresponding device, a calibration method and a computer program product.

Description

[0001] Method and device for inspecting a reflective surface

[0002] The invention relates to an optical method of a reflective surface of an object arranged in a measuring space and a corresponding device which has at least one 2-dimensional illumination pattern for reflection on the reflective surface and a plurality of cameras for pixel-by-pixel acquisition of image information of the illumination pattern reflected on the surface.

[0003] The well-known method of deflectometry is often used for defect detection on or optical measurement of reflective surfaces. In deflectometry, an image of a known pattern reflected in the surface is captured with a camera. In known deflectometry methods, the pattern is designed in such a way that, for example, image processing allows for clear identification and localization of positions within the pattern, thus allowing evaluation in relation to the surface shape or possible defects.

[0004] One example of such a well-known method uses the so-called random dot pattern. A random dot pattern is a pattern of small, randomly distributed dots. With such a pattern, a point can be identified using the random dot pattern surrounding it. It is a typical single-image method because a single image is sufficient to identify the point. However, to identify a point, its immediate surroundings are needed, which must be imaged with sufficient quality via the reflective surface in order to still be able to be identified. This is not always the case with reflective images, especially when the local surface curvature causes the pattern to be excessively distorted. This happens very easily with more curved or wavy surfaces. In extreme cases, the image can even be reversed.Therefore, multi-image methods are used instead. In a multi-image method, the coding of the point is not in its surroundings but in the sequence of gray values ​​that a pixel displays in several different patterns. One such widespread method is the so-called phase-shift method. In this method, stripe patterns with sinusoidal gray value curves in different phase positions are used as a pattern. Depending on the measurement task, at least three different phase positions and two or more stripe directions are required for each point on the object's surface. This requires a complex computational process for determining the shape of the reflective surface.

[0005] In the known deflectometry methods described above, both the camera and the pattern are calibrated, i.e., their relative positions in space are known. From this, the surface can be reconstructed. In order to find small errors (defects) on the surface, it is often necessary to analyze not the reconstructed surface, but its curvature field. If a pattern point can be clearly identified in the camera image, it is known which pattern point is reflected at the corresponding point on the surface, and the position of this pattern point is clearly determined in space, since both the position of the pattern in space and the position of the point within the pattern are known. Furthermore, the position of the corresponding image point in the camera is clearly determined in space, because both the imaging parameters of the camera and their position in space are known.This also means that a straight line in space is known on which the surface point lies in which the sample point currently being viewed is reflected. However, this does not yet determine the surface point itself. Any point on the straight line could be the desired surface point (also known as the ambiguity problem of deflectometry). However, each point has a specific (but unknown) surface normal because the mirror condition "angle of incidence equals angle of reflection" must be met at that point. A second condition is therefore required to determine the point precisely. This is often solved using so-called stereo deflectometry. In this case, a second camera looks at the same area of ​​the surface. This allows the ambiguity to be resolved. However, this determination is very complex because the correspondences in the two cameras cannot be established from the images but must first be searched for with great effort.This is because two cameras looking at the same point on the surface from different positions will not see exactly the same reflected point on the pattern. Another difficulty with this method is that, for curved surfaces, it is often difficult to cover the entire surface with two cameras.

[0006] The object of the present invention is therefore to provide a method or device that reduces or avoids the above problems. The object is also to provide a corresponding calibration method.

[0007] The above object is achieved by a method having the features of claim 1, by a device having the features of claim 7, by a calibration method having the features of claim 13, a computer program product having the features of claim 14 and a corresponding computer-readable data carrier according to claim 15.

[0008] In particular, the above object is achieved by a method for inspecting a reflective surface of an object arranged in a measuring space with a device which has at least one 2-dimensional illumination pattern for reflection on the reflective surface and a plurality of cameras for pixel-by-pixel acquisition of image information of the illumination pattern reflected on the surface, wherein the position and orientation of the plurality of cameras and of the at least one 2-dimensional illumination pattern are known in a common coordinate system and the illumination pattern represents a time-varying illumination pattern, comprising the following steps:

[0009] • Providing a measurement sequence with N (N > 2) steps to be carried out one after the other, in which at each step each of the plurality of cameras records a section of the measurement space illuminated with the illumination pattern, so that each camera generates a total of N pieces of image information for each point of the section of the measurement space recorded by the respective camera, wherein the illumination pattern is changed in a predetermined manner at each step with respect to the other N-1 steps,

[0010] • Generation of ground state image information using the measurement sequence from the measurement room without an object and transmission of the ground state image information to an evaluation device,

[0011] • Generation of object image information by means of the measurement sequence when the object is arranged in the measurement space, and transmission of the object image information to an evaluation device,

[0012] • Processing the transmitted basic state image information and object image information by means of the evaluation device in such a way that segmentation information is determined therefrom, wherein the segmentation information is calculated from a comparison of the object image information with the basic state image information, and

[0013] • Processing the transmitted object image information by means of the evaluation device in such a way that output data is generated therefrom with regard to an object-related inspection task taking into account the determined segmentation information, wherein the inspection task comprises, for example, the generation of defect information about the object and / or the calculation of surface normals for a plurality of points in at least one section of the reflective surface of the object and / or the calculation of the curvature for a plurality of points in at least one section of the reflective surface of the object as output data.

[0014] The method according to the invention enables a less complex inspection of a reflective surface (i.e., a surface reflecting visible light) of an object using a plurality of cameras, wherein the inspection tasks can include a defect analysis in at least one section of the reflective surface and / or the determination of the shape of the reflective surface (represented by determining the surface normal or the curvature for a plurality of points of the reflective surface in at least one section of the reflective surface). For example, less computing time is required, and the inspection results are more accurate. This is achieved by incorporating, with a high probability, only the object image information generated by reflection at the reflective surface into the inspection with a view to determining the output data.This is advantageous because, when inspecting various objects with different reflective surfaces in the measurement room, the cameras cannot always cover the surface perfectly. In practice, the edges and other locations in the images captured by the cameras contain sections that do not contain reflections from the reflective surface. These sections can, for example, represent the illumination pattern itself, sections of the object's surface without reflection of the illumination pattern, or other background sections. Occlusions by neighboring cameras or by the sensor housing can also "disturb" the image information.Therefore, the invention provides that when determining the output data for the inspection task, only those camera pixels are included that are highly likely to actually capture image information generated by illumination pattern reflections on the object surface. These pixels can be identified through segmentation. In other words, by taking the segmentation information into account, those determined object image information that do not show illumination patterns reflected on the respective surface can be separated from the object image information that actually contains such reflection image information.

[0015] The inspection method can be implemented as a computer-implemented method, i.e., a method carried out using a data processing device (computer). This device has a processor and a memory unit readable by the processor. The data processing device comprises, in particular, the evaluation device and its functionality. In addition, the data processing device can control at least one lighting device that generates the at least one 2-dimensional lighting pattern, in particular the temporal sequence of the lighting signals and their changes, as well as the plurality of cameras, in particular the synchronization of the generation of the image information (i.e., the camera recordings) with the control of the at least one lighting device. Furthermore, the data processing device can effect the transmission of the image information from the cameras to the evaluation device.

[0016] In one embodiment, during segmentation, the normalized ground state image information and normalized object image information generated during the N measurement sequence steps are compared pixel by pixel for each camera of the plurality of cameras. Additionally, in another embodiment, a temporal change in the object image information generated via the respective measurement sequence (and possibly also the ground state image information) for each pixel of each camera can be evaluated with respect to the step-by-step change in the illumination pattern.Furthermore, as a result of the comparison and, if applicable, additionally as a result of the evaluation, each pixel (image point) of each camera is assigned a label that can assume exactly two or, if applicable, four values ​​(see explanations below). If one of the two label values ​​or, if applicable, at least one of the four label values ​​has been assigned to a pixel, the object image information generated for the respective camera pixel is taken into account for the respective inspection task. If the other of the two or, if applicable, four label values ​​is assigned, the object image information generated by the respective camera pixel is not taken into account. This allows the image information of the camera pixels to be assigned in a simple and comparatively low-computational-intensive way to determine whether they are suitable for evaluation with regard to the inspection task.The number and location of pixels whose image information is not included in the processing to generate the output data for the inspection task (evaluation) generally varies for each of the numerous cameras. In this case, it is also possible that the segmentation results in the object image information of all pixels of a camera being included in the evaluation. Conversely, it can also happen that only the object image information of a small portion of all pixels of the respective camera is included in the evaluation with regard to the inspection task. In this case, the basic state image information can be determined once for the respective measuring room (including the numerous cameras and the lighting pattern), as long as, for example, the cameras and the arrangement of the lighting pattern do not change.For inspection with a different / additional or modified camera or a new position of the illumination pattern, the basic state image information must be regenerated by the method or the corresponding device.

[0017] The inspection device, which is described in more detail below, can output the output data with regard to the inspection task at a corresponding output of the evaluation device, wherein the inspection task comprises, for example, the generation of defect information about the object (ie for at least one section of the object) and / or the calculation of surface normals for a plurality of points of at least one section of the reflective surface of the object and / or the calculation of the curvature for a plurality of points of at least one section of the reflective surface of the object as output data.

[0018] The defect analysis can include, as output data in the form of defect information, for example, the location of a defect, the defect type (scratch, dent, inclusion, orange peel, waviness, flatness, etc.), and / or the size of a defect. Alternatively or additionally, the output data includes, for example, the normal on the reflective surface of the object at each point (also referred to as a normal map) of the respective surface section or the curvature of the reflective surface of the object at each point (also referred to as a curvature map) of the respective surface section. A defect represents a deviation from a given shape (e.g., the shape of the reflective surface of the 3-dimensional model of the object), e.g., a scratch, a dent, an inclusion, orange peel, waviness (for a straight surface in the model), or flatness (for a curved surface in the model).The method according to the invention can be used to perform quality control of the object with the reflective surface. The reflective surface can comprise the entire visible light-reflecting surface or a part (i.e., a section) of the visible light-reflecting surface of the object. The object can be, for example, the glass cover of a mobile phone or painted vehicle attachments, such as rear-view mirror housings, antenna covers, fuel caps, B-pillars. The method for inspecting a reflective surface can also be applied to transparent objects. Typical applications include spectacle lenses, vehicle windshields, or other curved panes. The device used to inspect the reflective surface using the above-mentioned method is explained in more detail below.It has at least one 2-dimensional illumination pattern, which has a predetermined pattern with different brightnesses, as well as a plurality of cameras, which are arranged and oriented such that all cameras can capture the reflection of the pattern on the reflective surface or at least one predetermined section of the reflective surface to be inspected. The illumination pattern can also represent a 2-dimensional surface, in which all points / surface areas have the same brightness. The illumination pattern is temporally variable and, in one embodiment, can include the temporal change according to a periodic function, for example a sinusoidal function (sine function), a trapezoidal function, a rectangular function, a triangular function, or the like. This means that each point / smallest area unit of the illumination unit has a brightness that varies over time (e.g.according to a sine function or another function mentioned above). The temporal change can be achieved, for example, by a corresponding shift of a spatially variable (e.g. spatially periodic) pattern (e.g. a sine pattern) along the surface of the illumination pattern. In this case, all reflections of the pattern recorded by all cameras can cover the entire reflective surface of an object or only the specified section of the reflective surface of the object in which the shape of the surface and / or the defects present there are to be determined. Each of the plurality of cameras records a part of the reflective surface, i.e. a section of the reflective surface. With regard to the measuring space, the term "section" is used below (also referred to below as camera section). Each camera creates images of a section of the measuring space.The cameras are further arranged and oriented such that each point of the section to be inspected or of the entire reflective surface is observed by at least one camera of the plurality of cameras. In one embodiment, an overlap region on the reflective surface, which represents the transition from a first observed section of a first camera to a second observed section of a second camera, can be viewed by at least two cameras, for example adjacent cameras of the plurality of cameras. In other words, overlap regions arise where the first section observed by the first camera overlaps with the second section observed by the second camera. The image information captured by the cameras represents brightness data, with image information being generated for each pixel of each camera during a recording, i.e. a brightness value is captured.

[0019] Furthermore, an evaluation device is provided for processing the object image information and / or ground state image information transmitted by the cameras, for example a microprocessor or a computer with such a microprocessor, wherein the evaluation device can also be part of a data processing device described above. The evaluation device is electrically connected to each of the plurality of cameras. In one embodiment, the evaluation device can be connected to one or, if appropriate, to several screens, each of which generates the illumination pattern. The 2-dimensional illumination pattern generated, for example, as described above, is also calibrated in the common coordinate system.

[0020] The cameras are designed, for example, as CCD cameras with light-sensitive electronic components (pixels) arranged in a matrix arrangement. All of the cameras in the plurality of cameras generate a brightness value for each of their pixels in a 2-dimensional matrix. This brightness matrix embodies the object image information or the ground-state image information according to the respective measurement sequence, depending on the configuration of the measurement space in which the image information is generated. Each camera transmits the generated image information to the evaluation device after recording. In one embodiment, the electronic components can have different spectral sensitivities.

[0021] By calibrating to the common coordinate system, the viewing direction and location of each camera, and thus the location of each pixel, are known for each of the multiple cameras. Furthermore, the locations of the screen(s) for the respective illumination pattern are known. The at least one 2-dimensional illumination pattern can be generated using a suitably extended image, projection onto a screen, or using a monitor. The latter option is particularly flexible, as the illumination pattern can be easily changed and adapted to the conditions of the reflective surface to be examined. The respective pattern and its temporal change can be achieved simply by appropriately controlling the monitor points.As already explained above, it is necessary that the at least one illumination pattern is arranged and oriented such that all cameras can observe the reflection of at least part of this pattern(s) in the reflective surface. The at least one illumination pattern is also calibrated in the common coordinate system of the plurality of cameras. The illumination pattern can, for example, be a pattern with a sinusoidal brightness curve, e.g. in stripe form, and different patterns with a sinusoidal brightness curve with different amplitudes (light intensity) and / or different period lengths (or frequencies) can be used. These illumination patterns can run in different directions. It is also possible to use several different sinusoidal stripe patterns, for example running in different directions.The change in illumination for generating N (N > 2) pieces of image information per camera pixel in a measurement sequence can be achieved by a corresponding shift of the respective illumination pattern (e.g., in a direction perpendicular to the stripes of a striped illumination pattern) or a corresponding shift of the object. In one embodiment, the measurement sequence can be used for different patterns with a periodic brightness curve with different amplitudes and / or different period lengths.

[0022] For the plurality of cameras, the measurement sequence can, for example, include 4, 6, 8, or 12 captures of the illumination pattern reflected from the reflective surface of the object and each time changed, i.e., performing 4, 6, 8, or 12 capture steps per measurement sequence. If a periodic illumination pattern (1 period = 360 degrees) is used, e.g., a sinusoidal illumination pattern (other illumination patterns are listed below), this means that the illumination pattern is shifted relative to the reflective surface by 90, 60, 45, or 30 degrees of a period of the illumination pattern, respectively, for each step of the measurement sequence. The spatial shift simultaneously achieves the temporal change in the illumination pattern described above, in relation to the pixels of all cameras.

[0023] The measuring space represents the 3-dimensional space that can be captured by all of the plurality of cameras. Different object types can be arranged and inspected in the measuring space, i.e. examined with regard to the respective inspection task. The arrangement of the specific object of the respective object type in the measuring space is such that the reflective surface to be inspected or the section of the reflective surface to be inspected is arranged entirely within the measuring space, so that the entire surface or the section to be inspected is completely captured by the plurality of cameras. Accordingly, it is necessary that the at least one illumination pattern is arranged such that these reflections are generated along the entire reflective surface or a section thereof to be inspected.As already explained above, the inspection task can include defect detection and / or shape determination of the reflective surface (or a section thereof) of the object. The shape determination is carried out, for example, by means of the above-mentioned calculation of the curvature for each point of the object's reflective surface and / or the surface normal for each point of the object's reflective surface. Each object type represents an object for which a large number of identical specimens can be inspected, for example as part of quality control. In one embodiment, a single 3-dimensional digital model (e.g., CAD model) is available for each object type and stored in the evaluation device for use during inspection (see below).For example, an object type may represent a glass cover of a first mobile phone model or a glass cover of another mobile phone model or a windshield of a specific car model, etc.

[0024] If the objects of the respective object type are arranged in essentially the same position in the measurement space, the ground-state image information is generated only once for this measurement space by applying the previously explained measurement sequence and transmitted to the evaluation device. If the measurement space changes (cameras, position of the illumination pattern, possibly other contents of the measurement space such as the sensor housing), it is necessary to redetermine this ground-state image information.

[0025] The object image information is generated for each object of the respective object type to be measured by applying the same measurement sequence, and the resulting object image information is transmitted to the evaluation device. There, it is processed together with the ground-state image information, and output data relating to the respective inspection task is generated at an output of the evaluation device. Based on the segmentation information, only the object image information of the pixels of the plurality of cameras is included in the generation of the output data (evaluation), which with a high probability contain signals reflected from the reflective surface (e.g., the pixels to which the label with the first value is assigned - see below for details).

[0026] During segmentation, normalization can be achieved, for example, by projecting the brightness values ​​determined in a measurement sequence for the respective pixel (i.e., when generating the ground state image information or when generating the object image information) onto the same brightness value range (e.g., the range from 0 to 255). This eliminates disruptive influences of object color and ambient light on the evaluation (i.e., processing with regard to the inspection task) or segmentation.

[0027] The aforementioned comparison of the ground-state image information and the object image information, and the aforementioned, possibly additional, evaluation with regard to the gradual change in the illumination pattern, are performed separately for each pixel of each camera based on the values ​​determined from each measurement sequence. Information from neighboring pixels is not included.

[0028] In detail, during the comparison (also known as similarity analysis), for example, for each pixel of each camera, after the respective normalization, the N normalized ground state image information is subtracted individually and in the respective chronological order from the N normalized object image information. This means that, for example, for N=6, the ground state image information NG1, NG2, NG3, NG4, NG5, NG6 and object image information NO1, NO2, NO3, NO4, NO5, NO6 generated for a pixel in the order 1, 2, 3, 4, 5, 6 are subtracted as follows: NO1 - NG1; NO2 - NG2; etc. up to NO6 - NG6. Of the six difference values ​​determined in this way, their minimum is subtracted from their maximum. This difference is the so-called range value for the comparison (or similarity analysis).

[0029] The additional assessment performed in one embodiment with regard to the gradual change in the illumination pattern includes a comparison with the change in the illumination pattern over time. This analysis is also referred to as “pattern analysis.” It improves the accuracy of the segmentation. It is based on the assumption that if the respective image information is similar to the change in the illumination pattern, the pixel has captured either the illumination pattern itself or the reflected illumination pattern. If this similarity does not exist, the respective pixel of the respective camera “sees” a different location, for example a background. For example, the temporal change in the illumination pattern can include a sinusoidal change (i.e., follow a sine function, meaning that the brightness of the illumination pattern at each point changes analogously to a sine function).Alternatively, periodic rectangular, trapezoidal, or triangular functions can be used. Depending on the shape of the change in the illumination pattern, an operation is applied to the image information, which allows the assessment of the shape of the change in the signal recorded in the respective pixel. In the example of the sine function, the image information of a single measurement sequence can be subtracted from one another in reverse order, e.g., NO4-NO1, NO5-NO2, NOB-NO5, NO1-NO4, NO2-NO5, NO3-NO6 (for the object image information in the above example). The range of the pattern analysis of the object image information is determined by subtracting the minimum from the maximum of these differences. The range of the pattern analysis can also be determined analogously for the ground-state image information. Other operations can be used accordingly for other forms of change, depending on the respective form.For other functions of temporal change of illumination, other operations adapted to the respective function of the change can be used to assess the form of the change in the image information.

[0030] For a small range of pattern analysis of the object image information, the respective pixel "sees" the illumination pattern itself or a reflection of the illumination pattern on the object or other objects in the measurement space. In this case, the pattern analysis returns the value "TRUE," while for the remaining (larger) ranges, the value "FALSE."

[0031] If the similarity analysis range is small, it's likely that the pixel in question is seeing either a background or the illumination pattern itself. In this case, the similarity analysis returns a "TRUE" value; otherwise, it returns a "FALSE" value.

[0032] The ranges of the pattern analysis and, if applicable, the similarity analysis, for which the values ​​"TRUE" (ie 1) or "FALSE" (ie 0) are assigned, are defined in advance, whereby the ranges depend on the entire possible value range of the brightness values ​​of a pixel.

[0033] The segmentation can therefore be carried out based on the range values ​​determined as shown above for the example of a sine function explained here as a change function of the illumination signal according to the following table, and a corresponding label value can be assigned to the respective pixel on the basis of the specified range values:

[0034] The labels are determined based on the results of the similarity analysis and, if applicable, the pattern analysis, as shown in the table above. If only the similarity analysis is used, only the image information of those pixels with the label value FALSE (i.e., 0) is used for the evaluation with regard to the inspection task. It can also be assumed that pixels that see the object without an illumination pattern occur in comparatively small numbers due to the pose design of the cameras. To further exclude such pixels, the pattern analysis can be used.When using similarity and pattern analysis in combination, for example, only the image information of pixels with a label value of "FALSE" from the similarity analysis and a label value of "TRUE" from the pattern analysis can be used to determine the output data for the inspection task. Image information from pixels that have been assigned the other combinations of label values ​​contained in the table are not used to generate the output data.

[0035] With regard to the inspection task in the form of defect detection, this can be carried out, for example, by evaluating the light-dark transitions present in the illumination pattern or in its reflection on the reflective surface. For example, the shape of the light-dark transition, a contrast in a specified area of ​​the light-dark transition, and / or a change in brightness over a specified distance across the light-dark transition can be evaluated, and based on the evaluation, it can be determined whether a defect is present in the area in question of the reflective surface recorded by the respective camera. This is described in detail in the document DE 10 2021 123 880 A1, the content of which is hereby incorporated into the description by this reference. Alternatively or additionally, defects can be detected and classified with regard to their type of defect based on the image analysis of a large number of recordings.A method for such classification is described in detail in document DE 102023 111 989 A1. The content of this document is also incorporated into the description by reference to this publication.

[0036] In one embodiment, a result of defect detection is the position(s) of the defect in the section of the reflective surface observed by the respective camera and thus in the common coordinate system, wherein several determined defect positions can be combined to form one (extended) defect, for example, based on the fact that they are located directly adjacent on the reflective surface. In one embodiment, the defect information determined during defect detection can include a defect position in the common coordinate system, wherein, for example, the size of the defect is determined from the defect position via the assignment of the defect location using the 3-dimensional digital model (e.g., CAD model) of the object type, if necessary by means of 3-dimensional reconstruction or 2-dimensional correction.Here, the 3-dimensional digital model of the object is the same for all objects of a given object type. In this exemplary embodiment, based on knowledge of the position and orientation of the plurality of cameras, the fixed position of the object in the measuring space, and knowledge of the position of the defect in the section of the reflective surface observed by the respective camera, the position of the defect in the common coordinate system can be determined by the intersection point of the camera's line of sight with the 3-dimensional digital model of the object. This conversion can be performed for all points of a defect and for all defects, and the evaluation device can accordingly determine all defect positions in the 3-dimensional coordinate system and on the reflective surface, and thus on the object.From this, the size of the respective defect can be derived, taking into account the corresponding properties (curvature of the reflective surface) at the position of the defect. This is an example of a method for determining the size of a defect when the position of the inspected object in relation to its specified position in the measuring space does not deviate significantly from the specified position (ideal position) on which the calculation is based. Alternatively, to compensate for incorrect positioning of the object compared to the ideal position, the surface can be reconstructed from a starting position using the determined curvature or normal information with respect to the reflective surface or a section thereof and registered with the 3-dimensional digital model of the object.During this registration, the reconstructed surface is compared with the 3-dimensional digital model of the object, and based on the comparison, it is determined which section and / or via which combined rotation-Z-translation displacement of this model the section of the reconstructed surface corresponds. This can be done, for example, using ICP (Iterative Closest Point). In a further alternative, a 2-dimensional contour of the section captured by one of the plurality of cameras, which can be obtained from the curvature and / or normal information, can be registered with the 2-dimensional contours of the reflective surface of the 3-dimensional digital model, and from this an affine (i.e., affine in the geometric sense) or perspective correction matrix can be determined for this camera.This correction matrix also compensates for the fact that the real object is not positioned exactly at the specified location where the 3-dimensional digital model is intended in the measuring space. This improves the accuracy of determining the 3-dimensional coordinates of the identified defects in the common coordinate system. The backprojection of the detected defects onto the 3-dimensional digital model (CAD model), as described above, enables a precise determination of the defect size. For each individual camera pixel, the metric area it covers in the digital model is determined. If the defect consists of several connected pixels, the areas for all connected pixels are combined accordingly.This is advantageous because the image scale can vary for different cameras and the viewing angle of each individual ray of view depends on the position and shape of the object at that position.

[0037] In one embodiment, the inspection task includes calculating surface normals for a plurality of points in at least one section of the reflective surface of the object and / or calculating the curvature for a plurality of points in at least one section of the reflective surface of the object. The respective section can comprise part of the reflective surface or the entire reflective surface of the object. In this case, for example, starting from a predetermined starting point (support point) on the reflective surface, the surface height and / or the surface angle as well as the normal to the surface are precisely determined. In one embodiment, a single starting point on the reflective surface is sufficient, for which the surface height and / or the surface angle as well as the normal to the surface are precisely determined.The starting point can be precisely determined, for example, using stereo deflectometry (deflectometry using two cameras that view the starting point on the reflective surface from different directions). Then, starting from this starting point, the normal at a point on the reflective surface adjacent to the starting point can be estimated based on the object image information, and so on. Once the normals to the respective points on the reflective surface have been determined for a large number of points, a corresponding normal field (also called a normal map) has been determined for the respective section. Accordingly, a curvature field (curvature map) can be determined from the normal field (normal map) for the respective section by differentiating the normals at the corresponding points.In one embodiment, the curvature at each point on the reflective surface can be determined from the normal map using the operation of mathematical derivation. Since a plurality of cameras are used to capture the reflective surface, the normal map and / or the curvature map for a larger section of the reflective surface or the entire reflective surface can be compiled from the normal / curvature results of the individual cameras. In the overlapping areas of the respective cameras, measurement results from two (or more than two) cameras are included in the determination of the normal and / or curvature. In particular, outlier normals and / or outlier curvatures can be identified during the evaluation if necessary.Outlier normals or outlier curvatures are characterized by the fact that they contain a large change compared to all neighboring normals or curvatures, and the change is limited to a single point on the reflective surface. In one embodiment, the at least one outlier normal in the normal map or outlier curvature is identified by comparing it with normals or curvatures of neighboring points on the reflective surface and / or by analyzing the local contrast in the respective image information. This provides a particularly simple way to prevent artifacts in the normal map or curvature map in advance.Once the outlier normals or outlier curvatures have been identified, they can be corrected in one embodiment by taking into account the object image information from another (different) camera for the respective point on the reflective surface and deriving the respective normal / curvature at this point from this. This smooths out steps / jumps in the normal or curvature map that arise due to the optical properties of the cameras. In a further embodiment, the curvature or surface normals of neighboring points can be integrated to determine the corrected curvature or the corrected surface normal at a point on the reflective surface. Neighboring points can be understood not only as the areas of the surface immediately adjacent to the respective point, but also as points further away.In one embodiment of the method, the steps of identifying the at least one outlier normal or outlier curvature and generating the corrected normal or the corrected curvature are repeated iteratively until a predefined termination criterion is reached. The iterative repetition increases the accuracy of the normal map or the curvature map. A termination criterion can, for example, represent the number of iterations. Alternatively or additionally, the corrected normals or the corrected curvature of the current iteration step can be compared with those of the previous iteration step. If the changes in the corrected normals or the corrected curvatures between these two iteration steps are below a predefined threshold, the iteration can be terminated.

[0038] In one embodiment of the method, the normal of a point on the reflective surface is given greater weight when generating the normal map or the curvature map if a section corresponding to the recording section is observed by at least two cameras. This means that the normal of the camera is preferentially used for the normal map (or curvature of the camera for the curvature map) that was determined from object image information that lies in the center of the area captured by the respective camera (and not in the edge area). This trade-off can be made because in this case at least two cameras are observing the same point on the reflective surface. This measure also increases the accuracy of the normal map orthe curvature map is improved, since in the area of ​​the center of the recording section of a camera, ie in the area of ​​the optical axis, the imaging errors of the respective camera are small compared to those of the edge areas.

[0039] In one embodiment with regard to the inspection task of generating defect information about the object, a value for the orange peel characteristic is determined for a predetermined section of the reflective surface of the object. The orange peel characteristic contains information about how pronounced the orange peel is. This can be calculated by the evaluation device, for example, as the standard deviation of the curvatures of the reflective surface determined within a predetermined measuring spot. With a large standard deviation, the orange peel has a large amplitude, whereas with a small standard deviation, the orange peel has a small amplitude. In one embodiment, low-frequency components can be subtracted before determining the standard deviation. The measuring spot is a section of the surface of the object in which the orange peel characteristic is to be determined.When determining the orange peel characteristic, the accuracy of this value can be improved by determining a camera-specific correction value for each camera in advance using a 3-dimensional digital orange peel model. The 3-dimensional digital orange peel model represents a test object that is assumed to have the same degree of orange peel throughout. Using the model, values ​​for local standard deviations are determined for the individual cameras or sub-areas of the observation sections, and corresponding correction factors are derived from the various values. The orange peel characteristic is another example of output data related to the inspection task.

[0040] The above object is further achieved by a device for inspecting a reflective surface of an object arranged in a measuring space, which device has at least one 2-dimensional illumination pattern for reflection on the reflective surface and a plurality of cameras for pixel-by-pixel acquisition of image information of the illumination pattern reflected on the surface and an evaluation device, wherein the position and orientation of the plurality of cameras and of the at least one 2-dimensional illumination pattern are known in a common coordinate system and the illumination pattern represents a time-varying illumination pattern, wherein the device is configured such that

[0041] • a measurement sequence with N (N > 2) steps to be carried out one after the other is provided, in which at each step each of the plurality of cameras records a section of the measuring space illuminated with the illumination pattern N times, so that each camera generates a total of N pieces of image information for each point of the section of the measuring space recorded by the respective camera, wherein the illumination pattern is changed in a predetermined manner at each step with respect to the other N-1 steps,

[0042] • Ground state image information is generated by means of the measurement sequence from the measurement room without an object and the generated ground state image information is transmitted to the evaluation device,

[0043] • Object image information is generated by means of the measurement sequence when the object is arranged in the measurement space, and the generated object image information is transmitted to the evaluation device,

[0044] • the transmitted basic state image information and object image information are processed by the evaluation device in such a way that segmentation information is determined therefrom, wherein the segmentation information is calculated from a comparison of the object image information with the basic state image information, and

[0045] • the transmitted object image information is processed by the evaluation device in such a way that output data is generated with regard to an object-related inspection task and taking into account the determined segmentation information, wherein the inspection task comprises, for example, the generation of defect information about the object and / or the calculation of surface normals for a plurality of points in at least one section of the reflective surface of the object and / or the calculation of the curvature for a plurality of points in at least one section of the reflective surface of the object.

[0046] The device has the advantages explained above in connection with the corresponding method. Corresponding embodiments of the method are also conceivable. Therefore, reference is made to the above explanations regarding the inspection method.

[0047] In one embodiment, the device is configured such that during segmentation, for each camera of the plurality of cameras, the normalized ground state image information and normalized object image information generated in the N measurement sequence steps are compared pixel by pixel, wherein in one embodiment, a temporal change in the object image information generated via the respective measurement sequence (and in one embodiment, additionally ground state image information) of each camera is evaluated in each case with regard to the step-by-step change in the illumination pattern, wherein, as a result of the comparison and optionally in addition to the evaluation, each pixel of each camera is assigned a label that can assume exactly two or, optionally, four values, wherein, if one of the two or, optionally, at least one of the four label values ​​has been assigned to a pixel,the object image information generated for the respective camera pixel is taken into account for the respective inspection task and when assigning the other of the two or, if applicable, four label values, the respective generated object image information is not taken into account.

[0048] In one embodiment, the device is configured such that the defect information includes a defect position in the common coordinate system, wherein, for example, the size of the defect is determined from the defect position via the assignment of the defect location using a 3-dimensional digital model of the object, optionally by means of 3-dimensional reconstruction or 2-dimensional correction.

[0049] In one embodiment, the device is configured such that a value for the orange peel characteristic is determined for a predetermined section of the reflective surface of the object, wherein a camera-related correction value for determining the orange peel characteristic value is determined in advance for each camera using an orange peel model.

[0050] In one embodiment, the device is configured such that at least one individual curvature value of a surface point and / or at least one individual normal value of a surface point is corrected, if necessary by taking into account object image information from at least two cameras relating to the surface point.

[0051] In one embodiment, the device is configured such that the correction is carried out on the basis of curvature values ​​and / or normal values ​​of neighboring points on the surface of the object and / or an analysis of the local contrast of a section of the surface comprising the surface point.

[0052] Furthermore, the above object is achieved by a method for calibrating a device specified above, in which the position and orientation (calibration) of the plurality of cameras in the common coordinate system is determined using a 3-dimensional cube with markers. This enables a particularly simple, fast, and accurate calibration of the plurality of cameras.

[0053] The aforementioned data processing device comprises a processor, which is a functional module that interprets and executes instructions / commands of algorithms, and comprises an instruction control unit, an arithmetic unit, and a logic unit. The processor may comprise at least one of a microprocessor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA - a digital integrated circuit into which a logic circuit can be programmed), a discrete logic circuit, and any combination of these components. The data processing unit may also comprise a memory unit, an input module (e.g., keyboard or touchpad), a power supply module (e.g., battery), and a display module (e.g., display).The data processing unit can be implemented as a real hardware resource, such as a smartphone, desktop computer, server, notebook, cluster / warehouse-scale computer, embedded system, or the like, or as a virtualized computer resource. Furthermore, the data processing unit can include a transmitter / receiver (transceiver) for exchanging data with the 3D scanner. The data processing device also includes an interface for exchanging data with the at least one lighting device, the plurality of cameras, and an output unit (e.g., a screen) for outputting the output data with respect to an object-related inspection task.

[0054] As already explained above, the methods explained above can each be implemented, for example, as a computer program or computer-implemented method comprising instructions which, when executed, cause a processor of the data processing unit and the elements of the device to carry out the steps of the above method, wherein the computer program includes a combination of the steps described above and data definitions which enable the computer hardware to carry out computing or control functions, and / or which represents a syntactical unit which conforms to the rules of a specific programming language and which consists of declarations and statements or instructions which are required for the functions, tasks or problem solutions explained above.Accordingly, the above object is also achieved by a computer program product comprising instructions which cause the above-specified device to carry out the corresponding steps of the above-specified methods.

[0055] Accordingly, a computer program product is disclosed comprising instructions that, when executed by the respective processor of the respective data processing unit, cause the respective device to perform the steps of the methods defined above. Accordingly, a computer-readable medium storing such a computer program product is disclosed. The computer program product may be a software routine.

[0056] Further advantages, features, and possible applications of the present invention will become apparent from the following description of exemplary embodiments and the drawings. All described and / or illustrated features, individually or in any combination, constitute the subject matter of the invention, regardless of their summary in the claims or their references.

[0057] They show schematically:

[0058] Fig. 1 shows the basic arrangement of the components of the device for inspecting a reflective surface,

[0059] Fig. 2 shows a first embodiment of the inspection device according to the invention with the object arranged in the measuring space in a side view,

[0060] Fig. 3 shows a first possibility for dividing the cameras of the device according to Fig. 2 into subgroups for calibration in a side view,

[0061] Fig. 4 shows a second variant for dividing the cameras of the device according to Fig. 2 into subgroups for calibration in a side view,

[0062] Fig. 5 - 7 the calculation of the coordinate transformations for the division of the cameras into sub-groups according to Fig. 3 in a view from the side,

[0063] Fig. 8 - 10 the calculation of the coordinate transformations for the division of the cameras into sub-groups according to Fig. 4 in a view from the side,

[0064] Fig. 11 Height map of the object according to Fig. 2, calculated as the average of the height maps determined separately for each camera without outlier correction, where the plotted heights are in the range of - 0.2 mm and 0.2 mm, Fig. 12 the height map of the object according to Fig. 2 with the proposed method with outlier correction, where the plotted heights are in the range of - 0.2 mm and 0.2 mm,

[0065] Fig. 13 the curvature map of a reflective surface of the object according to Fig. 2 without outlier correction, where the plotted curvatures are in the range of - 10 m' 1 (dpt) and 10 rrr 1 (dpt) lie,

[0066] Fig. 14 the curvature map of the reflecting surface of the object according to Fig. 2 with outlier correction, where the curvatures shown are in the range of - 10 m' 1 (dpt) and 10 rrr 1 (dpt) lie,

[0067] Fig. 15 shows another possibility for the calibration of a device according to the invention in a perspective view from the side, as well as

[0068] Fig. 16 shows a further embodiment of a device according to the invention in a perspective view from the side.

[0069] Fig. 1 shows the basic structure of a device for determining the shape of a reflective surface 100 of an object using deflectometry. The device has a camera 103 that observes the illumination pattern 101 across the reflective surface 100. The illumination pattern 101 is displayed on a screen 102. The camera 104 sees a distorted pattern—the image information 104—corresponding to the reflection of the illumination pattern 101 on the surface 100.

[0070] Fig. 2 shows an embodiment of the device according to the invention, which uses and further develops the basic principle shown in Fig. 1. The embodiment shown in Fig. 2 has a plurality of cameras 203, for example twelve cameras 203 (e.g. CCD cameras), which observe the reflection of the illumination pattern 201 displayed on the screen 202 on the reflective surface 200 of an object, for example the cover glass for a mobile phone. The object is arranged within the measuring space, wherein the measuring space is formed by the 3-dimensional space spanned within the cameras 203. The cameras 203 are arranged and oriented such that each camera 203 observes a section of the reflective surface 200. Each camera generates image information from a section of the measuring space and in particular from the surface 200 of the object.In overlapping areas of these sections, at least two cameras 203 view the same section of the reflective surface 200. Each camera 203 thus generates image information 204, which is shown as an example for six cameras 203 on the left and right sides of Fig. 2. The arrow 206 assigns the captured image information 204 to the respective camera 203. In addition, the device has an evaluation device 207 in the form of a computer, which is connected to the cameras 203 and the screen 202. The evaluation device 207 processes the image information 204, which is continuously transmitted to it from the cameras 203. In addition, the evaluation device can control the imaging of the illumination pattern 201 on the screen 202 and the corresponding temporal synchronization of the camera recordings by means of the cameras 203.

[0071] In order to determine the shape of the reflective surface 200, the relationship between the cameras 203 and the illumination pattern 201 of the screen 202 must be known. For this purpose, a common coordinate system of the cameras 203 and the illumination pattern 201 displayed on the screen 202 is determined by means of a calibration. For the calibration, in one embodiment, the twelve cameras 203 can be divided into three subgroups. In a first embodiment, shown in Fig. 3, the left four cameras are divided into a left subgroup (framed together with screen 202 by means of a frame 209), the center four cameras are divided into a center subgroup, and the right four cameras are divided into a right subgroup. For each of these subgroups, a calibration is carried out in a separate coordinate system by means of resection.For each calibration, the screen 202 is in the same position, and the same pattern is displayed on the screen 202. Thus, the position of the illumination pattern is the same for all calibrations of the subgroups. Furthermore, instead of the reflective surface to be examined, a known flat mirror 210 with points, each with a known, defined position, is used for the calibration.

[0072] Alternatively, the division into subgroups can be carried out as shown in Fig. 4. Here, cameras 1 to 6 form the left subgroup, cameras 5 to 8 the central subgroup, and cameras 7 to 12 the right subgroup. In this exemplary embodiment, the left and central subgroups of cameras have two common cameras (cameras 5 and 6), and the right and central subgroups also have two common cameras (cameras 7 and 8). In the exemplary embodiment shown in Fig. 4, the calibration is also initially carried out separately for each subgroup on the flat mirror with the defined points in each of its own coordinate systems.

[0073] The calibration in the own coordinate system of each subgroup is carried out, for example, using the procedure described above on the flat mirror with points in a defined position.

[0074] A camera subgroup is then defined as a master subgroup with a master coordinate system. In the two exemplary embodiments of subgroup definition shown in Figs. 3 and 4, the central subgroup is selected as the master subgroup with the master coordinate system. A transformation matrix is ​​then determined as a coordinate transformation, which converts the coordinate system of the left subgroup into that of the central subgroup and the coordinate system of the right subgroup into that of the central subgroup of cameras. This is easily possible because in the subgroup subdivision in Fig. 3, each subgroup has the same pattern position and in the subgroup subdivision in Fig. 4, the left and right subgroups each have two cameras in common with the central subgroup. The calculation of the respective transformation matrices is shown in Figs. 5 to 7 for the exemplary embodiment in Fig. 3 using the arrows 212 and in Fig.8 to 10 for the embodiment of Fig. 4 using the arrows 214.

[0075] The determined transformation matrices are then used to transform all cameras and the illumination pattern 201 displayed on the screen 202 into a common coordinate system, for example, the coordinate system of the central subgroup. This allows the cameras and the screen pattern to be easily calibrated into a common (global) coordinate system.

[0076] Alternatively, a 3-dimensional cube 310 with a reflective surface featuring markings can be used for calibration. Such a cube, arranged in the measurement space between five cameras 303, is shown in Fig. 15. The cube 310 has a reflective surface on each of its five upper sides, each provided with markings (e.g., dots). The illumination pattern is not shown in this figure, but is used for calibration analogously to the above method.

[0077] With the 3-dimensional cube 310, it is possible to calibrate all cameras 303 simultaneously in the common coordinate system if the position of all used side surfaces of the cube 310 in this coordinate system is known. The cube 310 is recorded in different positions (can be achieved, for example, by rotating the cube by less than 90° around the vertical axis). Such a calibration can be carried out significantly faster than with the above method. In particular, for the intrinsic calibration of each camera, the cube is recorded in different positions, while for determining the extrinsic parameters, i.e. for calibrating the position of the cameras 303 relative to one another in the common coordinate system, a single position of the cube 310 is sufficient.

[0078] The common coordinate system is then used to inspect the surface 200, ie to determine the shape of the specular surface 200 using a normal map or a curvature map and / or to detect defects and, if necessary, to determine their size.

[0079] The well-known phase-evaluation method of deflectometry is used here. For example, for the inspection of a reflective surface 200, a stripe pattern with a sinusoidal brightness gradient can be used as the illumination pattern, the reflection of which on the reflective surface 200 is recorded twelve times with a phase shift by each camera 203. The phase shift between consecutive images is 30 degrees of one period of the stripe pattern. This recording sequence with the 12 images and the associated illumination pattern setting corresponds to the measurement sequence described above. All images from all cameras 203 in a measurement sequence of the reflective surface 200 of the object represent the object image information. The spatial shift of the sinusoidal stripe pattern achieves a temporal change in the illumination for each pixel of each camera.Furthermore, the measurement sequence is carried out in advance using the same calibrated cameras without the object located in the measurement space. The resulting images from all cameras provide the baseline image information used for segmentation.

[0080] The illumination pattern 201 is displayed on a screen 202 (see Fig. 2). Alternatively, as shown in Fig. 16, for inspection with a plurality of cameras 403, multiple illumination patterns can be used, which are generated on three screens 402 and radiated in the direction of the reflective surface of the object arranged in the measuring space. In this exemplary embodiment of a device according to the invention, three screens 402 are used, which are arranged next to one another and inclined to one another. This allows for better inspection of reflective surfaces that have angled sections. All three illumination patterns are also calibrated in the common coordinate system.

[0081] Before the generated object image information is processed with regard to the respective inspection task, the segmentation described above is performed. This determines for each pixel of each camera 203 whether or not it captures the illumination pattern reflected from the surface 200. As can be seen in Fig. 2, the reflected illumination patterns each form only a small area 208 of the image information 204 of the respective camera section.

[0082] As already explained above, the segmentation means that hidden areas (e.g. by another camera, the sensor housing) that are contained in the image information - see the black dots in the two upper image information items in Fig. 2 and marked with the reference number 217 - are not taken into account when determining the normal map or the curvature map. Background areas outside the reflective surface or areas in which a camera directly observes the illumination pattern are also not included in the evaluation of the object image information with regard to the inspection task. This can be achieved, for example, by assigning a corresponding label to corresponding pixels of the respective camera - as described above. Either the similarity analysis alone can be used for segmentation or a combination of similarity analysis and pattern analysis.Both have been presented above, so reference is made to the above explanations, which describe the procedure specifically for an illumination pattern containing a striped sinusoidal pattern. An analogous procedure, adapted to the respective illumination pattern, can be applied to other illumination patterns.

[0083] Subsequently, based on the image information 204 determined by each camera 203, surface normals are determined at each point on the reflective surface 200, i.e., the normal map. Only those image information of the pixels of each camera 203 are included in the evaluation for which it was determined, based on the segmentation, that these most likely contain reflections of the illumination pattern of the screen 202 on the surface 200 of the object. Furthermore, a starting point is assumed, the position of which on the reflective surface 200 is determined, for example, using stereo deflectometry. The starting point is, for example, in the center of the reflective surface. Furthermore, the values ​​of the normal map are composed of the normal values ​​of the individual cameras. Such a map is shown in Fig. 11.Figure 11 shows that steps (outliers) occur in the transition areas between the values ​​of individual cameras, the cause of which was explained above. These are identified and corrected as shown above, whereby the identification and correction can be repeated iteratively. The normal map resulting from such an iteration is shown in Figure 12. The procedure for determining a curvature map for the reflective surface is analogous. Figure 13 shows a curvature map composed of the values ​​determined from the individual cameras 203, which has artifacts (outliers) (blue areas). The curvature map is determined and composed as a mathematical derivation from the normal values ​​of the individual cameras. Outliers can also be identified and corrected in the curvature map, as already described above.The identification and correction of outliers can be repeated iteratively until a termination criterion specified above is reached. The result of such an iteration is the calculation of the curvature map shown in Fig. 14.

[0084] It has also been explained above that when determining the surface normal for a specific point on the reflecting surface, the result (image information) of a camera that is central in relation to the entire image information of this camera (or close to its optical axis) can be given greater weight.

[0085] For defect detection, which can be performed alternatively or in addition to the above shape determination, the object image information generated by the cameras 203 can also be used. In this inspection task, only those image information from the pixels of a camera that are highly likely to capture the image information of the reflective surface (with reflected illumination pattern) are included in the evaluation. The defects can be detected as described above and categorized according to their type (scratch, dent, inclusion, orange peel, waviness, flatness, or the like). The position of the defect in the common coordinate system can also be determined.By linking the determined position data of the defect with a 3-dimensional digital model of the object (CAD model), the size of the defect in relation to the reflective surface can also be determined, as described in more detail above.

[0086] From the results of the exemplary embodiment of the method according to the invention, which are shown in Figs. 12 and 14, and the above explanations, it is evident that the method is capable of determining the shape of the reflective surface over a large curvature range, e.g., curved edges, as well as detecting defects in this surface. The method according to the invention generates complete and continuous normal and curvature maps of a reflective surface, as well as corresponding defect information, in a short time and in a single measurement. The method is fast, robust, reliable, and achieves accurate results. Furthermore, a simple and fast calibration method is proposed.

Claims

Patent claims:

1. A method for inspecting a reflective surface (200) of an object arranged in a measuring space, comprising a device having at least one 2-dimensional illumination pattern (201) for reflection on the reflective surface (200) and a plurality of cameras (203) for pixel-by-pixel acquisition of image information of the illumination pattern (201) reflected on the surface (200), wherein the position and orientation of the plurality of cameras (203) and of the at least one 2-dimensional illumination pattern (201) are known in a common coordinate system, and the illumination pattern represents a time-varying illumination pattern, comprising the following steps: • Providing a measurement sequence with N (N > 2) steps to be carried out one after the other, in which at each step each of the plurality of cameras records a section of the measurement space illuminated with the illumination pattern, so that each camera generates a total of N pieces of image information for each point of the section of the measurement space recorded by the respective camera, wherein the illumination pattern is changed in a predetermined manner at each step with respect to the other N-1 steps, • Generation of ground state image information by means of the measurement sequence from the measurement space without an object and transmission of the ground state image information to an evaluation device (207), • Generation of object image information by means of the measurement sequence when the object is arranged in the measurement space, and transmission of the object image information to an evaluation device (207), • Processing of the transmitted basic state image information and object image information by means of the evaluation device in such a way that Segmentation information is determined, whereby the segmentation information is calculated from a comparison of the object image information with the ground state image information, and • Processing the transmitted object image information by means of the evaluation device in such a way that output data is generated therefrom with regard to an object-related inspection task taking into account the determined segmentation information, wherein the inspection task comprises, for example, the generation of defect information about the object and / or the calculation of surface normals for a plurality of points in at least one section of the reflective surface of the object and / or the calculation of the curvature for a plurality of points in at least one section of the reflective surface of the object as output data.

2. Method according to claim 1, characterized in that during the segmentation for each camera of the plurality of cameras, the normalized ground state image information and normalized object image information generated in the N measurement sequence steps are compared pixel by pixel, wherein, for example, a temporal change in the object image information generated via the respective measurement sequence of each camera can be evaluated in each case with regard to the step-by-step change in the illumination pattern, wherein as a result of the comparison and optionally additionally of the evaluation, each pixel of each camera is assigned a label which can assume exactly two or optionally four values, wherein if one of the two or optionally at least one of the four label values has been assigned to a pixel,the object image information generated for the respective camera pixel is taken into account for the respective inspection task and when assigning the other of the two or, if applicable, four label values, the object image information generated for the respective camera pixel is not taken into account.

3. Method according to one of the preceding claims, characterized in that the defect information includes a defect position in the common coordinate system, wherein from the defect position, for example, the size of the defect is determined via the assignment of the defect location by means of a 3-dimensional digital model of the object, if necessary by means of 3-dimensional reconstruction or 2-dimensional correction.

4. Method according to one of the preceding claims, characterized in that a value for the orange peel characteristic is determined for a predetermined section of the reflective surface of the object, wherein for each camera a camera-related correction value for determining the orange peel characteristic value is determined in advance on the basis of an orange peel model.

5. Method according to one of the preceding claims, characterized in that at least one curvature value of a surface point and / or at least one normal value of a surface point is corrected, if necessary by taking into account object image information from at least two cameras relating to the surface point.

6. Method according to claim 5, characterized in that the correction is carried out on the basis of curvature values and / or normal values of neighboring points of the surface of the object and / or an analysis of the local contrast of a section of the surface comprising the surface point.

7. Device for inspecting a reflective surface (200) of an object arranged in a measuring space, which has at least one 2-dimensional illumination pattern (201) for reflection on the reflective surface (200) and a plurality of cameras (203) for pixel-by-pixel acquisition of image information of the illumination pattern (201) reflected on the surface (200) and an evaluation device (207), wherein the position and orientation of the plurality of cameras (203) and of the at least one 2-dimensional illumination pattern (201) are known in a common coordinate system and the illumination pattern represents a time-varying illumination pattern, wherein the device is configured such that • a measurement sequence with N (N > 2) steps to be carried out one after the other is provided, in which at each step each of the plurality of cameras records a section of the measuring space illuminated with the illumination pattern N times, so that each camera generates a total of N pieces of image information for each point of the section of the measuring space recorded by the respective camera, wherein the illumination pattern is changed in a predetermined manner at each step with respect to the other N-1 steps, • Ground state image information is generated by means of the measurement sequence from the measurement space without an object and the generated ground state image information is transmitted to the evaluation device (207), • object image information is generated by means of the measurement sequence when the object is arranged in the measurement space, and the generated object image information is transmitted to the evaluation device (207), • the transmitted basic state image information and object image information are processed by the evaluation device (207) in such a way that segmentation information is determined therefrom, wherein the segmentation information is calculated from a comparison of the object image information with the basic state image information, and • the transmitted object image information is processed by the evaluation device (207) in such a way that output data with regard to a object-related inspection task and taking into account the determined segmentation information, wherein the inspection task comprises, for example, the generation of defect information about the object and / or the calculation of surface normals for a plurality of points in at least one section of the reflective surface of the object and / or the calculation of the curvature for a plurality of points in at least one section of the reflective surface of the object as output data.

8. Device according to claim 7, which is set up such that during segmentation for each camera of the plurality of cameras, the normalized ground state image information and normalized object image information generated in the N measurement sequence steps are compared pixel by pixel, wherein, for example, a temporal change in the object image information generated via the respective measurement sequence of each camera can be evaluated in each case with regard to the step-by-step change in the illumination pattern, wherein as a result of the comparison and optionally additionally of the evaluation, each pixel of each camera is assigned a label which can assume exactly two or optionally four values, wherein if one of the two or optionally at least one of the four label values has been assigned to a pixel,the object image information generated for the respective camera pixel is taken into account for the respective inspection task and when assigning the other of the two or, if applicable, four label values, the respective generated object image information is not taken into account.

9. Device according to one of claims 7 to 8, which is set up in such a way that the defect information contains a defect position in the common coordinate system, wherein from the defect position, for example, the size of the defect can be determined via the assignment of the defect location by means of a 3-dimensional digital model of the object, if necessary by means of 3-dimensional reconstruction or 2-dimensional correction.

10. Device according to one of claims 7 to 9, which is set up in such a way that a value for the orange peel characteristic is determined for a predetermined section of the reflective surface of the object, wherein for each camera a camera-related correction value for the determination of the orange peel characteristic value is determined in advance on the basis of an orange peel model.

11. Device according to one of claims 7 to 10, which is set up such that at least one individual curvature value of a surface point and / or at least one individual normal value of a surface point is corrected, if necessary, by taking into account object image information from at least two cameras relating to the surface point.

12. Device according to claim 11, characterized in that the correction is carried out on the basis of curvature values and / or normal values of neighboring points of the surface of the object and / or an analysis of the local contrast of a section of the surface comprising the surface point.

13. A method for calibrating a device according to one of claims 7 to 12, characterized in that the determination of the position and orientation (calibration) of the plurality of cameras (203) in the common coordinate system is carried out by means of a 3-dimensional cube with marks.

14. A computer program product comprising instructions which cause the device according to claims 7 to 12 to perform the corresponding steps of the method according to any one of claims 1 to 6.

15. A computer-readable medium storing a computer program product according to claim 14.

Images