METHOD AND FLATBED TOOL MACHINE FOR DETECTING A SUSPENSION POSITION OF A SUPPORT BED

Description

[0001] The present invention relates to a method for detecting the suspension position of a support bar in a pallet intended for use with a flatbed machine tool, in particular a laser cutting or plasma cutting flatbed machine tool. Furthermore, the invention relates to a flatbed machine tool suitable for carrying out the method.

[0002] Flatbed machine tools, particularly laser cutting or plasma cutting flatbed machine tools, typically have a machine area in which the (e.g., laser) processing takes place. Upstream and optionally downstream storage areas for the raw material and / or the cut material are assigned to the machine area. With flatbed machine tools, material sheets to be processed (e.g., sheet metal) can be fed to the machine area on a pallet. A pallet typically comprises an arrangement of support bars, by means of which the material sheet to be processed can be stored in a support plane. Depending on the design, support bars can only be inserted at certain positions on the pallet.

[0003] The pallet can, for example, be part of a pallet changer, which represents a storage area upstream of the flatbed machine tool. After machining, a large number of cut workpieces can be present on the pallet. The workpieces can be removed from the pallet manually by an operator or (semi-)automatically. The relative position between the manufactured workpiece and the supporting structures on the pallet is important for both the production process and the removal process.

[0004] The support webs are usually plate-shaped and have evenly spaced support web tips that are formed along an edge / side of the support web. The support webs are inserted vertically with the support web tips pointing upwards in order to form numerous support points that define a support plane for a sheet of material and the cut workpieces. When inserted, several support webs run parallel to one another. The support webs are arranged so that their spacing from one another is adjustable along a longitudinal extent of the pallet and can, for example, be suspended from a frame of the pallet at equally spaced (hanging) positions. This makes it possible to configure the arrangement of the support webs on a pallet as required, i.e. to select the positions of the support webs along the longitudinal extent, for example, specifically for a machining job.In the area of ​​the support web tips, support areas are formed in which the support webs can be in contact with a material sheet stored in the support plane.

[0005] Furthermore, modern nesting approaches allow the allocation of workpieces to be cut out on a material sheet to be processed, depending on the existing support web configuration. In particular, the workpieces are aligned with regard to the position of the support web tips or the support areas. By adapting the sheet allocation to the position of the support web tips, wear on the support web tips can be reduced and process quality can be increased, for example by preventing the tilting of workpieces and the subsequent collision of the tilted workpiece with a cutting head of the flatbed machine tool, or by moving to specific (gentle) piercing positions for the laser. The starting point for a sheet allocation adapted to the position of the support web tips or the support areas is knowledge of the support web configuration intended for use, which includes, among other things:by the mounting positions of the support bars. However, since it is not known which support bars have been set up, not all applications of the flatbed machine tool may be able to take the support bar positions into account.

[0006] It is known to detect the position of the support bars, for example, using a distance sensor (e.g., attached to the cutting head) within the machine area or by laser triangulation. However, such methods involve unproductive downtime due to the "scanning" of the empty pallet.

[0007] The article "Measuring the supporting slats of laser cutting machines using laser triangulation", by F. STRUCKMEIER ET AL, and published in THE INTERNATIONAL JOURNAL OF ADVANCED MANUFACTURING TECHNOLOGY, discloses the preamble of claims 1 and 16.

[0008] From DE 20 2020 101 810 U1 a device for determining the actual condition of support strips of a workpiece support is known.

[0009] A laser engraving system is known from CN 204 867 817 U.

[0010] EP 3 338 946 A1 discloses a method for measuring and compensating thermal expansion in a machine.

[0011] One aspect of this disclosure is based on the object of specifying a method for detecting the suspension position of a support bar and / or a support bar tip. The method should be able to be carried out, in particular, even under adverse conditions frequently encountered in practice, for example, in an undefined lighting situation or in the case of wear and slagging of the support bar tips. The method should, if possible, deliver robust, fast, and as accurate results as possible despite the surface conditions of the support bars.

[0012] At least one of these objects is achieved by a method according to claim 1 and by a flatbed machine tool according to claim 16. Further developments of the invention are specified in the subclaims.

[0013] In one aspect of the invention (see claim 1), a method is defined with which a suspension position of a support bar in a pallet can be detected. The pallet is equipped for use with a flatbed machine tool, in particular a laser cutting or plasma cutting flatbed machine tool, with a plurality of support bars along a longitudinal direction, wherein each of the support bars has a plurality of support bar tips along a main extension direction oriented transversely to the longitudinal direction of the pallet, and the support bar tips of the plurality of support bars define a support plane. The method comprises the steps: Creating a two-dimensional contrast image data set of the palette comprising a plurality of image pixels, wherein each image pixel is assigned a pixel value and a pixel area unit of the support plane and the contrast image data set comprises regions assigned to the support web tips as local pixel value extrema in a uniform pixel value background, determining a longitudinal position in the contrast image data set of those regions assigned to the support web tips of one of the plurality of support webs using the local pixel value extrema and deriving the insertion position of the support web in the palette based on the longitudinal position in the contrast image data set and an extent of the pixel area unit in the longitudinal direction.

[0014] In a further aspect of the invention (see claim 16), a flatbed machine tool is defined, in particular a laser cutting or plasma cutting flatbed machine tool: a pallet with a plurality of support webs suspended at suspension positions, each of the support webs having a plurality of support web tips defining a support plane, a camera for generating a two-dimensional camera image or a two-dimensional dark field camera image of the pallet and an evaluation device.

[0015] The camera and / or the evaluation device are designed for a Creating a two-dimensional contrast image data set of the palette comprising a plurality of image pixels, wherein each image pixel is assigned a pixel value and a pixel area unit of the support plane and the contrast image data set comprises regions assigned to the support web tips as local pixel value extrema in a uniform pixel value background.

[0016] The evaluation device is further designed for: Determining a longitudinal position in the contrast image data set of those regions associated with the support ridge tips of one of the plurality of support ridges using the local pixel value extrema and deriving the insertion position of the support ridge in the palette based on the longitudinal position in the contrast image data set and an extent of the pixel area unit in the longitudinal direction.

[0017] In some further developments, one of the regions assigned to the support web tips can be assigned to a support web tip and / or a flank leading to the support web tip.

[0018] In some embodiments, the uniform pixel value background may not represent a structure and / or may comprise image pixels that have a constant pixel value or a random pixel value distribution around a constant pixel value.

[0019] In some further developments, the creation of the contrast image data set of the palette may further include: Generating a two-dimensional camera image of the pallet in the visible and / or infrared spectrum, wherein the camera image has the image pixels arranged in rows and columns, each of which is assigned the pixel area unit of the support plane, Fourier transforming the camera image into a frequency domain and outputting a transform frequency data set, filtering the transform frequency data set to emphasize frequencies in the frequency domain that belong to the areas assigned to the support web tips, and inverse transforming the filtered transform frequency data set by means of inverse Fourier transform into an inverse transformed image data set and outputting the inverse transformed image data set as the contrast image data set.

[0020] In some developments, adjacent support web peaks can have a substantially equal peak spacing from one another in the main extension direction. Filtering the transformation frequency data set can then comprise identifying a frequency band in the frequency domain associated with the peak spacing and restricting the transformation frequency data set to the frequency band. The frequency band can, for example, be formed around a peak repetition frequency associated with the main extension direction. It can, in particular, result from: h ∗ Δx d , with h: number of pixels of the area of ​​the support plane shown in the camera image in the main extension direction, Δx: extent of the pixel area unit in the main extension direction and d: tip distance of adjacent support web tips.

[0021] In some developments, the two-dimensional camera image can be generated by generating a dark-field camera image of the pallet in the near-infrared frequency spectrum, and the dark-field camera image can be recorded from a recording direction that encloses an angle with the support plane that is in the range of 10° to 70°, in particular greater than 45°, for example 60°, and that extends at an angle in a range of 70° to 110°, in particular perpendicular, to the main extension direction.

[0022] In some embodiments, creating the contrast image dataset of the pallet may further comprise generating the contrast image dataset by generating a dark-field camera image of the pallet in the near-infrared frequency spectrum from a recording direction that forms an angle with the support plane. The angle may, for example, be in the range of 10° to 70°, in particular greater than 45°, for example 60°. The recording direction may extend at an angle in a range of 70° to 110°, in particular perpendicular, to the main extension direction.

[0023] In some embodiments, creating the contrast image data set of the pallet may further comprise illuminating the pallet from above with light of a near-infrared frequency spectrum from one or more illumination directions. The light may, in particular, be in the frequency spectrum from 845 nm to 855 nm and / or from 935 nm to 945 nm.

[0024] In some developments, at least one of the illumination directions can enclose an angle with the support plane which is less than 30° and in particular lies in a range from 10° to 20°, for example at approximately 15°, and the at least one of the illumination directions projected into the support plane can enclose an angle with a main extension direction of the support web which is less than 30° and in particular lies in a range from 10° to 25°, for example at approximately 15°.

[0025] In some developments, generating the two-dimensional camera image or generating the two-dimensional dark-field camera image may comprise recording a plurality of partial camera images, each of which represents a partial area of ​​the pallet in two dimensions, and composing the partial camera images to form the two-dimensional camera image or the two-dimensional dark-field camera image of the pallet.

[0026] In some embodiments, determining the longitudinal position in the contrast image data set may comprise performing a template-matching image analysis using a template. The template may be embodied as an image data set of a support web having a plurality of support web tips, in particular of a support web arranged in an interior region of the pallet.

[0027] In some embodiments, determining the longitudinal position in the contrast image data set may comprise summing the pixel values ​​of the image pixels of the contrast image data set row by row according to the main extension direction and outputting a first distribution of first pixel value sums and determining a local extremum in the first distribution and outputting the row of the local extremum as the longitudinal position.

[0028] In some training courses, the procedure may also include: Summing the pixel values ​​of the image pixels of the contrast image data set column by column according to the longitudinal direction in the region of the local extremum and outputting a second distribution of second pixel value sums, determining a plurality of local extrema in the second distribution and outputting the columns of the local extrema and determining image positions of the regions assigned to the support web tips in the contrast image data set based on the corresponding row of the determined local extrema and the columns of the determined local extrema.

[0029] In some developments, the support webs can be suspended at predetermined suspension positions on a longitudinal side of the pallet, and the main extension direction can extend transversely to the longitudinal side, and the pallet can have a reference structure that is identifiable in the contrast image dataset as a reference pixel arrangement. The suspension position of the support web in the pallet can be derived by determining a distance between the reference pixel arrangement and the longitudinal position transversely to the main extension direction in the contrast image dataset and by identifying one of the predetermined suspension positions for the longitudinal position based on the determined distance.

[0030] In some embodiments, the pixel value can be a gray value, a color value, or a brightness value.

[0031] In some training courses, the camera can be set up to to generate the two-dimensional camera image of the pallet in the visible or infrared spectrum, wherein the camera image has image pixels arranged in rows and columns, each of which is assigned a pixel area unit of the support plane, or to generate a contrast image data set by recording the two-dimensional dark-field camera image of the pallet in the near-infrared frequency spectrum from a recording direction.In some developments, the evaluation device can be designed to carry out the method and / or can comprise a processor which is configured to Fourier transform the camera image into a frequency domain and to output a transformation frequency data set, to filter the transformation frequency data set to emphasize frequencies in the frequency domain which belong to the regions assigned to the support web tips, and to inversely transform the filtered transformation frequency data set by means of inverse Fourier transformation and to output the inversely transformed image data set as the contrast image data set.

[0032] In some developments, at least one lighting device can be provided, which is configured to illuminate the pallet with light of a near-infrared frequency spectrum from one of the lighting directions. At least one of the lighting directions can enclose an angle with the support plane that is less than 30° and in particular lies in a range from 10° to 20°, for example approximately 15°. Projected into the support plane, the at least one of the lighting directions can enclose an angle with a main extension direction of the support web that is less than 30° and in particular lies in a range from 10° to 25°, for example approximately 15°.

[0033] In some developments, the method can comprise determining individual positions of the regions assigned to the individual support web tips within the columns or rows assigned to the first local pixel value sum extrema. For this purpose, for example, a parallel projection image analysis can be carried out in which the pixel values ​​of the image pixels of the columns or rows of the contrast image dataset assigned to the first local pixel value sum extrema, i.e., which are assigned to a support web, are summed (row by row or column by column) perpendicular to the main extension direction of the support web to form a second pixel value sum assigned to the respective row or line. Based on this, the rows or columns whose assigned pixel value sums form a second local pixel value sum extrema can be determined.

[0034] In some further developments, determining the longitudinal position of the regions associated with the support bridge tips in the contrast image dataset may alternatively or additionally comprise performing a sliding window approach image analysis in which a framework is created around a template matching filter, or an image analysis based on a SIFT algorithm or an image analysis based on neural networks.

[0035] Concepts are disclosed herein that allow aspects of the prior art to be improved, at least in part. In particular, further features and their usefulness will become apparent from the following description of embodiments with reference to the figures. The figures show: Fig. 1 a schematic spatial representation of a flatbed machine tool with a pallet equipped with several support webs, Fig. 2 a flow chart to explain the method for detecting a suspension position of a support web in a pallet, Fig. 3 a rectified camera image of an exemplary pallet (bird's eye view), Fig. 4 a representation of an amplitude spectrum of a transformation frequency data set of the camera image related to a frequency domain, Fig. 5 juxtaposed sections of an exemplary contrast image data set generated by means of filtered Fourier transformation and the associated camera image of a pallet, Fig. 6 a flow chart to explain the calculation of the contrast image data set of the pallet using Fourier transformation and filtering in the frequency domain, Fig. 7 an exemplary contrast image data set generated by means of dark field illumination, Fig.Fig. 8 shows a flowchart explaining the imaging of a contrast image dataset of the palette using dark-field illumination and dark-field recording, and Fig. 9 shows a section of an exemplary template for a template image analysis.

[0036] The aspects described herein are based in part on the finding that the detection of the suspension position of a support bar using image recognition, particularly image recognition focused on detecting the support bar tips, can be performed robustly, quickly, and with high accuracy if a contrast image dataset is used in which the support bar tips are highlighted and other, "interfering" structures are largely suppressed or not displayed. Interfering structures can be caused, for example, by an undefined lighting situation, wear, slagging, and the surface condition of the support bar. Using such a low-interference contrast image dataset, incorrect support bar detections can be reduced. Furthermore, the image analysis effort can be reduced.

[0037] Furthermore, the inventors have recognized that a contrast image dataset can be created using either Fourier transformation or dark-field illumination or dark-field imaging. Furthermore, both methods can also be combined, for example, by subsequently filtering a (camera) image created using dark-field illumination or dark-field imaging using Fourier transformation before converting it into a contrast image dataset. Such a combination of both methods can enable an even more noise-free contrast image dataset.

[0038] In both cases, a camera first takes a picture of the (usually empty) pallet, which is then transformed into a bird's-eye view so that the support bars appear in the image lengthwise or crosswise, e.g., as (vertical or horizontal) lines extending along the main direction of the support bars. The image can also be composed of several partial images from different cameras and taken both outside ("in front of") and inside the machine (i.e., in the production area).

[0039] In the image to be evaluated, the support plane is displayed without perspective distortion in the contrast image dataset. For rectification, an originally perspective-distorted image can be processed, for example, using conventional homography methods. In this way, each image pixel can be assigned a pixel area unit of the support plane, particularly a square one, of the same size for all image pixels, for example 1 mm². With this rectification, objects of the same size in the support plane can be displayed as the same size in the contrast image dataset, regardless of the position of the individual object in the support plane or in the contrast image dataset. If the dimensions of an image pixel in a direction of extension of the support ridges and transversely thereto are known, the rectification can facilitate the determination of the longitudinal position of the regions assigned to the support ridge tips (and thus the determination of the longitudinal position of the associated support ridge).

[0040] The concepts disclosed herein are therefore based, among other things, on the use of a contrast image dataset of a pallet in which areas associated with the support web tips are highlighted. For example, such high-contrast image datasets can be obtained using special imaging arrangements or image analysis methods (or a combination of both methods). As examples, filtering as part of Fourier image processing and the application of dark-field imaging are explained in more detail herein. Filtering using Fourier image processing and the application of dark-field imaging can be used alternatively or combined with one another, for example by first capturing a (camera) image of the pallet using dark-field imaging, which is then processed by filtering using Fourier image processing.Modifications to this or further approaches will become apparent from the combined description of these exemplary embodiments.

[0041] The Fourier transform is a transformation commonly used in image processing of an image data set related to a first reference system, the so-called image or position space, into a transform frequency data set related to a second reference system, the so-called frequency space.

[0042] In the Fourier transformation, a periodically regular pattern in spatial space, i.e., related to the geometry of the palette, is assigned a frequency in the frequency domain of the transformation image data set. Periodically regular patterns in image space along the pixel alignment directions are represented in the representation of the transformation frequency data set in an amplitude spectrum as horizontal or vertical (frequency) lines and are thus easily recognizable. The support ridge peaks also form a repeating pattern in the image data set along a main extension direction of the support ridge and can thus be assigned frequencies in the transformation frequency data set. These frequencies are also represented as lines in the amplitude spectrum (assuming appropriate alignment of the support ridges).

[0043] By filtering the transformation frequency data set, the frequencies assigned to the support ridge peaks can be emphasized, while the other frequencies can be (relatively) attenuated or even removed. By subsequently retransforming the filtered transformation frequency data set, an image data set (related to the image space) can be output in which the support ridge peaks are highlighted with high contrast. In contrast, "disturbing" structures or textures that are not clearly assigned to the support ridges are at least attenuated, so that, in other words, the areas of the support ridge peaks can be represented as local pixel value extrema in a uniform pixel value background within a contrast image data set of the palette. A local pixel value extrema can be a group of neighboring image pixels or individual image pixels.If the local pixel value extreme is formed as a group of neighboring image pixels, the image pixels belonging to the group each have significantly higher pixel values ​​than those image pixels surrounding the image pixel group of the local pixel value extreme and forming the uniform pixel value background. The pixel value can, for example, be a gray value, where a low gray value corresponds to a dark image pixel, and a high gray value to a bright (illuminated) image pixel.

[0044] A contrast image dataset of the pallet can also be obtained using dark-field illumination or dark-field imaging (dark-field strategy). Dark-field strategies are used, for example, in automated visual inspection. Dark-field strategies utilize a dark-field illumination device and a dark-field camera that records a dark-field image. The illumination is applied at one (often very shallow) angle, whereas the image can be recorded at a different (steep) angle. The majority of the light emitted by the dark-field illumination device is not reflected back to the dark-field camera. In the visual inspection described above, only defects in the surface scatter / reflect light, whereas in the present case of an image of a pallet with support ribs, only the tips of the support ribs and the edge leading to the tips of the support ribs scatter / reflect light back to the dark-field camera.The result is a largely dark dark-field image (referred to herein as a uniform pixel value background), with only the support ridge peaks and flanks appearing significantly brighter than their surroundings (referred to herein as local pixel value extremes). Such a dark-field image is another example of a palette contrast image dataset, in which the areas associated with the support ridge peaks are represented as local pixel value extremes in a uniform pixel value background.

[0045] By using light from a near-infrared frequency spectrum for (dark-field) illumination, extraneous light influences from sunlight or artificial lighting can be reliably suppressed even without enclosure. Extraneous light influences from sunlight can be suppressed, in particular, by using near-infrared light in the frequency spectrum from 845 nm to 855 nm and / or from 935 nm to 945 nm for illumination, as this corresponds to the band gaps in the spectrum of sunlight on the Earth's surface. Capturing a dark-field camera image of the pallet in the near-infrared frequency spectrum, particularly in the aforementioned frequency spectra, can be achieved, in particular, by using an appropriate spectral filter.

[0046] The determination of the mounting position is based on an analysis of the contrast image data set. Various analysis methods are disclosed herein as examples.

[0047] Template-matching image analysis is a method for detecting objects in an image dataset based on a two-step approach. First, a template of the object to be detected is created, then the occurrence of the template in the image dataset is assessed using a similarity measure, e.g., the sum of absolute differences. Template-matching image analysis is described in detail in G. Cheng and J. Han, "A survey on object detection in optical remote sensing images," ISPRS Journal of Photogrammetry and Remote Sensing, vol. 117, pp. 11-28, 2016.

[0048] The template, which can be designed as an image data set of a support web, can be generated in particular by recording a camera image or a dark-field camera image in the near-infrared frequency spectrum of a new (not worn, not slagged) support web.

[0049] Since perspective effects can change the appearance of the support bar depending on its insertion position along the pallet's length, a support bar that has been inserted in the inner area of ​​the pallet's length can be selected as a template. Such a template can be used to detect support bars that are inserted further "forward" or further "backward."

[0050] Parallel projection image analysis is also a method for detecting objects that extend across a certain section of a row or column of an image dataset. In particular, parallel projection image analysis allows for very simple and rapid identification of overlays (whose main extension direction extends along a row or row of image pixels) in the image by summing the gray values ​​of the image pixels row by row (or column by column). Overlays form a maximum in this "signal distribution" generated by projecting / summing the pixel values ​​along the overlays. Parallel projection image analysis is described, for example, in P. Du et al., "Improved peak detection in mass spectrum by incorporating continuous wavelet transform-based pattern matching," Bioinformatics, vol. 22, no. 17, pp. 2059-2065, 2006.

[0051] Further methods for image analysis of the contrast image dataset include a sliding window approach image analysis (see e.g. M. Barva, et al., "Parallel integral projection transform for straight electrode localization in 3-D ultrasound images," IEEE transactions on ultrasonics, ferroelectrics, and frequency control, vol. 55, no. 7, pp. 1559-1569, 2008), a SIFT algorithm-based image analysis (see e.g. D.G. Lowe, "Object recognition from local scaleinvariant features." in IEEE International Conference on Computer Vision, vol. 99, no. 2, 1999, pp. 1150-1157) and an image analysis based on neural networks (see e.g. D. Erhan, et al., "Scalable object detection using deep neural networks," in IEEE Conference on Computer Vision and Pattern Recognition, 2014, pp. 2147-2154 and K. He, et al., "Mask r-cnn," in Proceedings of the IEEE International Conference on Computer Vision, 2017, pp. 2961-2969).

[0052] In general, misidentifications can be caused by, for example, highly reflective edges on the support bars or by cutting slugs. To subsequently rule out such misidentifications, identified support bar candidates can be checked. For example, it can be checked how strongly the gray values ​​of the image section assigned to the support bar candidate fluctuate in the direction of the support bar tips. If the fluctuation is above a specified threshold, it can be assumed that the support bar candidate was incorrectly identified and excluded. In this way, misidentifications can be quickly and easily avoided when using parallel projection image analysis.

[0053] The following is based on Fig. 1 a system with a flatbed machine tool is described. Fig. 2 summarizes the steps of a procedure for recording an attachment position in a flowchart. Figuren 3 bis 6 The generation of a contrast image dataset using filtered Fourier transformation is explained. Figuren 7 and 8 The generation of a contrast image dataset using a dark field approach is explained. Fig. 9 illustrates a template for image analysis.

[0054] Fig. 1 shows a flatbed machine tool 1 with a pallet 3, a camera 5, lighting devices 7, 7' (shown in dashed lines) and an evaluation device 11 comprising a processor 9 and connected to the camera 5.

[0055] Along a longitudinal side 3a of the pallet 3 (which defines the longitudinal extent), support webs 13 can be suspended in a frame 4 of the pallet 3 in several discrete suspension positions 15, wherein suspension generally includes any type of stationary arrangement and optionally fastening. In some suspension positions 15, in the example of the pallet of the Fig. 1 no support bars 13 are mounted. It can be seen that the support bars 13 are more densely arranged near the machine tool than at the far end 3b of the pallet.

[0056] The support webs 13 extend along a common main extension direction 13a (in Fig. 1 along end sides 3b of the pallet 3), are each plate-shaped and have a plurality of support web tips 17. The support web tips 17 form / define a common support plane 17a. In the main extension direction 13a, consecutive / adjacent support web tips 17 are arranged at a distance from each other by a tip spacing 17b. Typically, the tip spacing 17b is essentially the same for all support webs 13.

[0057] A material sheet (not shown) intended for processing with the flatbed machine tool 1 can be supported on the support webs 13 in the support plane 17a. The support web tips 17 are in contact with the material sheet, so that, for example, during material processing, the laser beam can also interact with the support web tips 17. In order to be able to take the positions of the support webs 13 into account, for example, when arranging workpieces to be cut in the material sheet, automated detection of the positions of the support webs 13 is preferably desirable.

[0058] To ensure correct positioning of the material sheet, the pallet 3 has a guide arranged on the frame 4 and shown schematically in Fig. 1 The design of the stop 19 allows it to be used as an optically detectable reference structure 19a in image processing.

[0059] The camera 5 is connected, for example, by a camera arm 5' to a production housing of the flatbed machine tool 1. By means of the camera 5, a two-dimensional, in particular monochrome or colored, camera image of the pallet 3 in the visible or infrared spectrum (see image data set 21 Fig. 3 ) or a darkfield camera image of Palette 3 in the near-infrared frequency spectrum (see contrast image data set 23' Fig. 7 ) are generated. The respective image is taken from a recording direction 5a from the side above onto the pallet 3. Due to the oblique recording direction 5a, an optical or digital rectification of the camera image can be carried out for simple image processing. The Figuren 3 and 7 The images shown are correspondingly rectified camera images. As already mentioned, such a "rectified" camera image of the support plane 17a can be provided, for example, using the homography method.

[0060] The (common) main extension direction 13a of the illustrated support webs 13 and the longitudinal direction of the illustrated pallet 3 running transversely thereto can also be used in relation to the image data set 21, the contrast image data set (23) and the dark field camera image or the contrast image data set (23') for corresponding image-related or image data set-related directional information.

[0061] The camera image and the dark-field camera image contain image pixels arranged in rows and columns, each of which is assigned a pixel value and a typically square, equally sized pixel area unit of the support plane. The pixel value is a gray value, with a low gray value corresponding to a dark pixel and a high gray value to a light pixel. Color resolution is not necessary for the aspects discussed here, as image analysis primarily relies on intensity values.

[0062] The receiving direction 5a forms an angle 5b with the support plane 17a. In the exemplary case of Fig. 1 The receiving direction 5a runs perpendicular to the main extension direction 13a of the support webs 13. The angle 5b is, for example, in the range of 10° to 70°; in particular, it can be greater than 45°, for example, 60°. Deviations from the perpendicular orientation with respect to the main extension direction are, for example, ±20°.

[0063] By means of the illumination device 7 (shown in dashed lines), the pallet 3 can optionally be additionally illuminated with infrared light, for example, with light in the near-infrared frequency spectrum from 845 nm to 855 nm and / or 935 nm to 945 nm, in an illumination direction 7a. The illumination direction 7a forms an angle 7b with the support plane 17a. The angle 7b is, for example, less than 30° and lies in particular in a range of 10° to 20°, for example, approximately 15°. The illumination device 7 illuminates the support webs 13 from "behind," i.e., it shines onto the rear sides of the support webs 13—relative to the camera 5. Accordingly, the illumination direction 7a projected into the support plane 17a (perpendicular to the support plane 17a) forms an angle 7c with the main extension direction 13a of the support webs 13. The angle 7c is, for example, less than 30° and lies in a range of 10° to 25°, for example, approximately 15°.

[0064] Further lighting directions can be provided with additional lighting devices; for example, an additional lighting device 7' is provided in Fig. 1 shown. Multiple lighting devices allow for more even illumination of the support bars. It is important to ensure that no lighting device shines directly onto the camera to avoid oversaturating the camera ("camera iris"). Additionally, camera 5 can be used in combination with a frequency filter that transmits a specifically selected spectral range.

[0065] The flatbed machine tool 1 is configured to enable a method for detecting the suspension positions 15 of the support webs 13. The necessary image processing calculations can be performed, for example, with the processor 9 of the evaluation device 11. Accordingly, the camera 5 and optionally the illumination device 7 are connected to the processor 9 / the evaluation device 11 for receiving data (image data sets) and for controlling the same (triggering the image capture or the illumination). The evaluation device 11 can be designed as a PC, computing node, or similar suitable hardware and, in particular, as part of the control unit that controls the flatbed machine tool 1. The evaluation device 11 can thus be designed as part of a higher-level or local control system of the flatbed machine tool or as a separate unit.The evaluation device 11 is particularly configured to perform / monitor the method for determining the mounting positions during real-time operation of the flatbed machine tool 1. For this purpose, the underlying computing system comprises, for example, digital processor systems with microprocessor circuits having data inputs and control outputs, as well as databases that operate according to computer-readable instructions stored on a computer-readable medium. The instructions include, for example, computer routines for Fourier transformations, filtering, and image analysis. The evaluation device 11 can provide high computing power for real-time support.It may further comprise a long-term (non-volatile) memory for storing the program instructions as well as a very fast short-term (volatile) memory for (temporarily) storing acquired data and data sets generated during the execution of the method.

[0066] The method for detecting the suspension position 15 of the support web relates to a pallet 3, which, as described above, is equipped with a plurality of support webs 13 along a longitudinal direction of the pallet 3. Each of the support webs 13 has a plurality of support web tips 17 along the main extension direction 13a, which is oriented transversely to the longitudinal direction. The support web tips 17 define the support plane 17a. As shown in Fig. 2 As shown, the method comprises the following steps, which can be implemented at least partially as a computer-implemented method: In a step 100, a two-dimensional contrast image data set of the palette 3 comprising a plurality of image pixels is created. In this case, each image pixel is assigned a pixel value and a pixel area unit of the support plane 17a, and the contrast image data set includes regions assigned to the support web tips 17 as local pixel value extremes in a uniform pixel value background.

[0067] In a step 200, a longitudinal position of those regions which are assigned to the support web tips 17 of one of the plurality of support webs 13 is determined in the contrast image data set based on the local pixel value extrema.

[0068] In a step 300, the suspension position 15 of the support web 13 in the pallet 3 is derived based on the longitudinal position in the contrast image data set and an extent of the pixel area unit in the longitudinal direction.

[0069] Two approaches are described below as examples for creating a contrast image dataset (see step 100).

[0070] First, in connection with the Figuren 3 bis 6 explains how to create a contrast image dataset 23 using Fourier transformation. Figuren 7 and 8The creation using dark-field illumination and dark-field imaging is described. Advantages of the Fourier transformation are its flexible application and integration into existing imaging systems, as it often allows for the use of readily available image data. In contrast to this computationally intensive implementation, the dark-field approach essentially requires an independent camera system with optional dedicated illumination. Data processing is significantly simplified with the dark-field approach. The common goal of both approaches is to provide a contrast image dataset in which the support ridge peaks 17 are highlighted as much as possible, represented in the examples as "the brightest possible areas" in an otherwise as black as possible image (or in a random pixel value distribution around a constant pixel value).

[0071] Both approaches start with an image of the pallet when it is stationary in the machine tool or on the pallet changer. Preferably, the pallet is also completely empty. This initial image is perspectively transformed so that the support bars appear line-like in one of the pixel alignment directions (vertical or horizontal in the image). The initial image can also be composed of multiple camera images.

[0072] Fig. 3 shows a two-dimensional camera image 21 of the pallet, generated by the camera 5 in the visible spectrum, with several support webs along the main extension direction 13a, which in Fig. 1 In some cases, one can also see areas of the pallet frame whose long sides 3a extend in a longitudinal direction (according to Fig. 1 marked as x-direction) and whose end sides 3b extend along the main extension direction 13a.

[0073] The camera image 21 has a plurality of image pixels, each of which is assigned a gray value. The camera image 21 was recorded obliquely from above, so that with increasing distance of the support webs from the camera 5, not only the upper edge but also increasingly the side surfaces of the support webs facing the camera are visible in the camera image 21. In other words, due to the perspective, the individual support webs 13 are displayed differently in the camera image 21 depending on their position in the support plane 17a, in particular the suspension position in the longitudinal direction: The support webs in the left edge area of ​​the camera image 21 appear almost in plan view due to the relatively steep recording angle and thus appear line-like (only the top side is perceived). The support webs in the right edge area of ​​the camera image 21 appear in a type of oblique view due to the relatively flat recording angle, whereby their plate-shaped side surfaces are perceived.

[0074] However, since the camera image 21 has been rectified, for example, using the above-mentioned method of homography, each image pixel is assigned a square, equally sized pixel area unit of the support plane 17a, so that the camera image 21 corresponds to a distortion-free top view (bird's eye view) in this respect, even if the image content deviates from this due to the oblique recording direction.

[0075] Furthermore, in the camera image 21 the Fig. 1 The reference structure 19a shown can be recognized as the reference pixel arrangement 19b. The reference pixel arrangement 19b can be identified using image analysis methods (e.g., template matching image analysis, see below).

[0076] Due to the lighting situation, the support plane 17a may be unevenly illuminated, which is reflected in some areas of the support webs 13 appearing significantly brighter than others. In general, a strongly structured grayscale distribution can be seen in the camera image 21.

[0077] Furthermore, the individual support webs 13 can be in different states with regard to wear, slagging, and surface condition. For example, individual support web tips can be partially melted and / or deformed.

[0078] As part of the image processing, the processor 9 Fourier transforms the camera image 21 related to a spatial space of the Fig. 3 into a transformation frequency data set 21a related to a frequency space.

[0079] Fig. 4 shows the transformation frequency data set 21a in a so-called amplitude frequency representation, in which the brightness of an individual pixel represents a measure of the magnitude of the amplitude (squared magnitude) of the respective frequency. The frequency axes (fx-axis and fy-axis) represent the spatial frequencies in the two orthogonal analysis directions (x-direction, y-direction). Fig. 4 the two frequency axes fx, fy each have their origin in the middle (coordinate origin).

[0080] Due to the nature of the Fourier transformation and the regular periodic arrangement of the support ridge tips 17 and the support ridges 13 relative to each other, distinct brighter areas with local brightness extremes (peaks) in the square of the amplitude are visible. For example, "horizontal" lines Lx in the fx direction and vertical lines Ly in the fy direction were Fig. 4 These lighter areas are due to the geometry of the support webs 13, in particular to the repeating structure of the support web tips 17 in the main extension direction 13a and / or to the regularity of the suspension positions 15 in the longitudinal direction.

[0081] Taking into account the possible geometry of the loaded support bars of the pallet, for example, a binary filtering of the transformation frequency data set 21a can be carried out.

[0082] The support web tips 17, which are spaced apart by a distance 17b, form a periodic, regularly repeating pattern in the camera image 21 (in spatial space) extending along the main extension direction 13a. A peak repetition frequency (a spatial frequency in the y-direction) can be assigned to this pattern in the transformation frequency data set 21a. The peak repetition frequency is defined by the two horizontal lines Ly1, which are closest to the coordinate origin and equally spaced from it. Fig. 4 The peak repetition frequency is given by: h ∗ Δx d , with h: number of pixels of the area of ​​the support plane 17a shown in the camera image 21 in the main extension direction 13a, Δx: dimension of the pixel area unit in the main extension direction 13a and d: tip spacing 17b of adjacent support web tips 17.

[0083] Example values ​​are: Δx = 1 mm, e.g. h = 1600 image pixels and d = 16 mm, resulting in a peak refresh rate of 100 image pixels.

[0084] To emphasize the peak repetition frequency, the transformation frequency data set 21a can be filtered, ie, the transformation frequency data set 21a is limited to a frequency band. A filtered transformation frequency data set extends, for example, in a fixed frequency bandwidth around the peak repetition frequency. Fig. 4 For example, two frequency bands along the lines Ly1 are selected.

[0085] In this context, the inventors have further recognized that the subsequent determination (step 200 in Fig. 2 ) the position of the areas assigned to the support bridge tips in the contrast image data set can be determined quickly and reliably if high-frequency, fx spatial frequencies, whose amplitude contribution can be neglected (in Fig. 4 illustrated as the frequencies to the left and right of the dashed lines S1, S2) are removed from the transform frequency data set 21a.

[0086] The thus filtered and limited transformation frequency data set 21a includes, for example, only the amplitude values ​​in the rectangles B, for example, by setting the amplitude contributions of the remaining frequencies to zero. In a representation analogous to Fig. 4 the areas outside the rectangles B would be shown in black.

[0087] Subsequently, the processor 9 transforms the filtered transform frequency data set 21a back into the spatial domain using inverse Fourier transformation.

[0088] Because the transformation frequency data set 21a has been limited to the frequency band around the peak repetition frequency and thus all frequencies outside the frequency band are suppressed (filtered), a contrast image data set 23 (related to the image space) can be created by the subsequent back-transformation of the filtered transformation frequency data set 21a (see Fig. 5 ). In the contrast image data set 23, the support ridge peaks 17 are highlighted, and other "disturbing" structures or textures are at least substantially suppressed. The contrast image data set 23 accordingly represents the areas of the support ridge peaks 17 as local pixel value extrema 23a in a uniform pixel value background 23b. This results in a back-transformed image data set, which can be output as the contrast image data set 23 for further image analysis.

[0089] Fig. 5 illustrates this "disturbance-reducing" effect by displaying a section (right side of the figure) from the two-dimensional camera image 21 of the palette according to Fig. 3 and juxtaposes a section (left side of the figure) of the contrast image data set 23 generated from it using Fourier transformation. Several well-structured linear arrays of bright areas can be seen, each corresponding to the support bar tips of a support bar. Note that in the section shown, the arrays are equally spaced from each other (i.e., every mounting position was used).

[0090] Fig. 6 illustrates step 100 of creating the contrast image data set 23 using Fourier transformation. The underlying image processing comprises the following steps, which can be implemented at least partially as a computer-implemented method: In a step 110', a two-dimensional camera image 21 of the pallet 3 is recorded. Part of the recording process is a rectification, which is performed such that the image pixels of the camera image 21, which are arranged in rows and columns, are each assigned the same pixel area unit of the support plane 17a. The rectification can, for example, be calculated using the processor 9 of the evaluation device 11 based on the image data of the initial image or can be performed at least partially optically.

[0091] In a step 120', the camera image 21 is Fourier transformed into a frequency domain and output as a transform frequency data set 21a.

[0092] In a step 130', the transformation frequency data set 21a is filtered to highlight frequencies in the frequency space that belong to the areas assigned to the support web tips 17.

[0093] In a step 140', a Fourier inverse transformation of the filtered transformation frequency data set 21a is carried out by means of inverse Fourier transformation, so that an inversely transformed image data set can be output as contrast image data set 23.

[0094] In the second approach described below, a contrast image dataset is created using a darkfield image, optionally with darkfield illumination. For this purpose, as described in Fig. 1 As shown, the pallet 3 is illuminated by one or more illumination devices 7, 7' with light of the near-infrared frequency spectrum from the illumination direction 7a.

[0095] The implementation of the dark field approach (dark field strategy) to create a contrast image dataset of pallet 3 can take into account that the space above pallet 3 should be available for loading and unloading processes. For example, as in Fig. 1 shown, the camera 5 is attached to a rear wall of the flatbed machine tool 1 via a camera arm 5'.

[0096] The lighting device 7 can then illuminate the support web tips 17 from the side and from above so that reflected and scattered light is deflected upwards towards the camera 5. Furthermore, the lighting device 7 should not be directed at too steep an angle from above onto the pallet, since otherwise objects (e.g. material scraps) located under the pallet 3 could be illuminated and become part of the image. Furthermore, the lighting device 7 should not illuminate the side surfaces of the support webs 13 facing the camera 5. The lighting device 7 can preferably be directed from the side, at a shallow angle from above and behind (i.e. from the side opposite the camera 5). The lighting device 7 can, for example, be positioned at a distance of 40 cm from the pallet.For an illumination device 7, the dark field illumination strikes support web tips 13 located further away from the illumination device 7 at a flatter angle.

[0097] It should also be noted that sometimes severe glare spots ("camera glare") can occur if light can pass directly from the lighting device 7 into the camera 5. This can be the case if the lighting device 7 is located on the side of the pallet 3 opposite the camera 5. To avoid this, the lighting device 7 can be provided with appropriate covers. As mentioned, several lighting devices 7, 7' can be provided for uniform illumination of the entire support plane of the pallet 3.

[0098] Fig. 7 shows an exemplary result of the dark-field recording in the form of a dark-field camera image of pallet 3, which was recorded in the near-infrared frequency spectrum from the recording direction 5a using camera 5. The dark-field camera image can be used directly (after rectification) as the contrast image data set 23' for further image analysis. Alternatively, it would also be possible to filter the dark-field camera image according to the method described above using a Fourier transformation and only use the filtered and back-transformed image data set as the contrast image data set as the basis for steps 200 and 300.

[0099] In the example shown, camera 5 was fitted with a narrowband filter with a maximum transmittance of over 90% in the near-infrared frequency spectrum. As shown in Fig. 7 In the contrast image data set 23' (essentially) only the support web tips 17 and some flanks leading to the support web tips 17 are recognizable as bright areas in an otherwise largely black image.

[0100] The flowchart of the Fig. 8 illustrates step 100 of creating the contrast image dataset using a dark-field camera image. The underlying image acquisition comprises the following steps, which can be implemented at least partially as a computer-implemented method: In a step 105" (optional), the pallet 3 is illuminated from above with light of a near-infrared frequency spectrum, in particular in the frequency spectrum from 845 nm to 855 nm and / or from 935 nm to 945 nm, from one or more illumination directions 7a. At least one of the illumination directions 7a can be at the angles 7a, 7b specified above with the support plane.

[0101] In a step 110", the contrast image data set 23' is generated by generating a dark-field camera image of the pallet 3 in the near-infrared frequency spectrum from a recording direction 5a. The dark-field recording can be carried out at the angles 5a, 5b specified above with the support plane and the main extension direction.

[0102] In other words, in a preferred embodiment, a flatbed machine tool can be equipped with near-infrared (NIR) illumination for a pallet used for processing, as well as a camera with a narrowband filter. The narrowband filter prevents the influence of ambient light. If the camera is installed on the housing of the flatbed machine tool and the illumination is positioned along the length of the pallet so that the support bars are illuminated obliquely from behind, a dark-field image is created in which the tips and flanks of the support bars are brightly displayed and the rest of the image is dark.

[0103] Once the contrast image dataset 23, 23' has been created, the positions of the regions assigned to the support web tips 17 in the contrast image dataset 23, 23' are determined using image analysis. This can be done, for example, using a parallel projection for image analysis of the contrast image dataset 23, 23' or with a template matching image analysis. Other more complex image analysis methods such as a sliding window approach image analysis, a SIFT algorithm-based image analysis, or an image analysis based on neural networks can also be used.

[0104] Schematically, in Fig. 7 The result of a parallel projection along the main extension direction 13a is indicated at the lower edge of the image. By summing the pixel values ​​in the y-direction (the gray values ​​of the image are added column by column in the direction of the main extension direction 13a), an intensity profile 31 is obtained in the x-direction, i.e., along the longitudinal extension of the palette 3. A plurality of maxima 31a can be seen in the intensity profile.

[0105] Highly reflective edges on the support webs or cutting slugs can lead to misidentifications. These can be detected, for example, by checking how strongly the gray values ​​fluctuate in the main extension direction 13a for a candidate (a corresponding area of ​​the intensity profile 31). For example, the gray values ​​in a region of the x-axis around the candidate (e.g., in a range of + / - 20 image pixels around a maximum) can be added along the x-axis (parallel projection perpendicular to the main extension direction). Alternatively or additionally, in order to quickly classify a candidate as a misidentification, the average of the gray value fluctuations along the y-axis can be calculated. If this average is below a predetermined threshold, it is a misidentification.

[0106] In addition, Fig. 7 clarifies that the support bars can be mounted at equal distances. If one of the mounting positions 15 coincides with an intensity maximum in the intensity profile, the corresponding mounting position is marked with a "1" (mounting position set up). If an intensity maximum is missing at a mounting position 15, it is marked with a "0" (no support bar 13 mounted). In the case of Fig. 7 that not every suspension position 15 was used, i.e. not every suspension position 15 was equipped with a support bar 13.

[0107] A similar intensity profile of a parallel projection perpendicular to the main extension direction 13a, which is limited to a narrower area around the support ridge tips (e.g., + / - 4 image pixels), can also be used to identify individual support ridge tips in the y-direction. For this purpose, the signal maxima (sum of the gray values) can again be determined.

[0108] This is also in Fig. 7 For a support bar 13, a parallel projection in the x-direction was made to determine the exact position of a support bar tip 17. A corresponding intensity profile 35 in the y-direction was recorded with some maxima 35a in Fig. 7 The x-position of the suspension position (of a maximum 31a) and the y-position of one of the maxima 35a in the intensity profile 35 result in the two-dimensional position of a corresponding support web tip 17. A support web tip 37 is shown as an example using the x-position and the y-position in Fig. 7 These positions of the support web tips 17 can, for example, be used as a basis for the cutting process by taking the positions of the support web tips 17 into account in the planned arrangement of workpieces to be cut.

[0109] The parallel projection approach presented in connection with the dark-field camera image can be applied analogously to image data set 23.

[0110] Fig. 9 shows a template 25 for a template-matching image analysis. The template 25 shows areas 27 with slightly elevated pixel values, which in the template belong to the support web peaks and optionally to the flanks leading to the support web peaks. Typically, a template can be extracted from a plurality of images. The template 25 can, for example, be obtained from an image data set of a support web 13 having a plurality of support web peaks 17. For example, it can be obtained from an interior region of the pallet 3, since the support webs 13 arranged there can be viewed as a mean value for all support webs 13. The template 25 can also be created using the Fourier forward transformation and Fourier back transformation, optionally with appropriate filtering.

[0111] As in Fig. 9 As further shown, a template can be used, for example, asymmetrically with respect to the regions 27 in order to take existing structural information into account and prevent the erroneous recognition of a "web dip" as a "web peak." This is particularly the case if the supporting web is viewed from the side in the initial image, i.e., does not form a line. An asymmetric template can then be selected such that a region without a Fourier bandpass filter response is required next to the detected signal. In the case of a web dip, the peaks would be visible there, so that only the supporting web peaks provide a high degree of agreement with the template 25.

[0112] In template matching image analysis, determining the positions of the regions assigned to the support ridge tips 17 in the contrast image data set 23, 23' involves performing a comparison operation using the template 25 with respect to comparably large sections of the contrast image data set 23, 23'. The result of determining the positions using the template 25 primarily yields the x-positions of the template, which lead to high levels of agreement with the contrast image data set 23, 23'. These positions then correspond to the insertion positions 15 of the support ridges 13. Depending on the quality of the template and the contrast strength of the template 25 and the contrast image data set 23, 23', the respective positions of the support ridge tips 17 along the main extension direction 13a can also be recorded.

[0113] In the template matching image analysis, the assignment of the contrast image data set 23, 23' to the real position of the possible mounting positions 15 in the pallet 3 must also be included for determining the mounting positions 15, as described in connection with Fig. 7 with regard to the mounting positions 15 has already been explained.

[0114] In other words, if the positions of the areas assigned to the support web tips 17 in the contrast image data set 23, 23' are determined based on the local pixel value extrema 23a (step 200), the corresponding mounting positions 15 of the support webs 13 in the palette 3 can be derived based on the positions of the areas assigned to the support web tips 17 in the contrast image data set 23, 23' and the pixel area unit (step 300).

[0115] This can be done, for example, for a support web 13 by first determining the number of image pixels between the reference pixel arrangement 19b (see Fig. 3) and an x-position of the areas assigned to the support web tips 17 of the support web 13 is determined transversely to the main extension direction 13a. Subsequently, from the determined number of image pixels, a corresponding distance between the reference structure 19a (assigned to the reference pixel arrangement 19b) and the relevant support web 13 is determined using known dimensions of the pixel area units assigned to the image pixels (e.g. 1 mm x 1 mm). Based on the determined distance (between the reference structure 19a and the relevant support web 13) and known distances between the reference structure 19a and the (discrete) possible suspension positions 15, a suspension position 15 can be assigned to the relevant support web 13 and thus the suspension positions 15 of the support web 13 in the pallet 3 can be determined.

[0116] Comparing the two approaches presented as examples for generating the contrast image dataset, the dark-field approach has the advantage of not being dependent on favorable ambient light. A further advantage is that the floor beneath the pallet 3 and the front of the support bars 13 are not illuminated. This leads to less interference in the image. Accordingly, the evaluation of whether a support bar is inserted at a suspension position 15 becomes more robust thanks to the dark-field image. Misclassifications hardly occur. The dark-field image also enables the evaluation of the positions of individual support bar tips 17 on individual images with an acceptable error rate. Furthermore, the process can be carried out quickly using both the dark-field approach and the Fourier approach, since the image for the dark-field / Fourier analysis can be taken in fractions of a second.

[0117] In the case of Fourier image analysis, the current pallet support bar occupancy can be extracted directly from the initial image without additional tools, such as specific dark-field illumination. In this case, too, unlike scanning methods, there is hardly any downtime, since, as mentioned above, the pallet is empty shortly before a new sheet is placed on it. Implementation of Fourier image analysis requires only a suitable data processing environment to calculate the described transformations. The calculations can be performed on the machine tool controller or outsourced to an external server. Accuracies of 96% and more have been achieved.

Claims

1. A method for detecting a suspension position (15) of a support bar (13) in a pallet (3) which is fitted with a plurality of support bars (13) along a longitudinal direction (x), the pallet being used together with a flatbed machine tool (1), in particular a laser-cutting or plasma-cutting flatbed machine tool, wherein each of the support bars (13) has a plurality of support bar tips (17) along a main direction of extension (13a) aligned laterally relative to the longitudinal direction of a pallet (3), and the support bar tips (17) of the plurality of support bars (13) define a support plane (17a), characterized by the steps: - creating (100) a two-dimensional contrast-image dataset (23, 23') of the pallet (3), wherein a pixel value and a pixel area unit of the support plane (17a) is assigned to each image pixel, and the contrast-image dataset (23, 23') comprises regions that are assigned to the support bar tips (17) as local pixel-value extremes (23a) in a uniform pixel-value background (23b), - determining (200) a longitudinal position in the contrast-image dataset (23, 23') of regions that are assigned to the support bar tips (17) of one of the plurality of support bars (13) using the local pixel-value extremes (23a) and - deriving (300) the suspension position (15) of the support bar (13) in the pallet (3) on the basis of the longitudinal position in the contrast-image dataset (23) and an extent of the pixel area unit in the longitudinal direction (x).

2. The method according to claim 1, wherein one of the regions assigned to the support bar tips (17) is assigned to a support bar tip (17) and / or to a flank leading to the support bar tip (17).

3. The method according to one of the preceding claims, wherein the uniform pixel-value background (23b) does not represent a structure and / or comprises image pixels that have a constant pixel value or a random pixel-value distribution about a constant pixel value.

4. The method according to one of the preceding claims, wherein the creation (100) of the contrast-image data set (23, 23') of the pallet (3) also comprises: - creating (110') a two-dimensional camera image (21) of the pallet (3) in the visible or infrared spectrum, wherein the camera image (21) has the image pixels arranged in rows and columns, one each of the pixel-area units of the support plane (17a) being assigned to said image pixels, respectively, - Fourier transforming (120') the camera image (21) into a frequency domain and outputting a transformation-frequency dataset (21a), - filtering (130') the transformation-frequency dataset (21a) in order to emphasize frequencies in the frequency domain that belong to the regions that are assigned to the support bar tips (17), and - back-transforming (140') the filtered transformation-frequency dataset into a back-transformed image dataset using an inverse Fourier transformation and outputting the back-transformed image dataset as the contrast-image dataset (23).

5. The method according to claim 4, wherein adjacent support bar tips (17) in the main direction of extension (13a) have substantially the same tip spacing (17b) from one another, and wherein the filtering (130') of the transformation-frequency dataset (21a) comprises: - identifying a frequency band assigned to the tip spacing (17b) in the frequency domain, wherein the frequency band is formed around a tip repetition frequency assigned to the main direction of extension and is obtained from: h * Δ x d , where h: is a number of pixels of the region of the support plane (17a) in the main direction of extension (13a) that is represented in the camera image (21), Δx: is the extent of the pixel area unit in the main direction of extension (13a) and d: is the tip spacing (17b) of adjacent support bar tips (17), and - restricting the transformation-frequency dataset (21a) to the frequency band.

6. The method according to claim 4 or 5, wherein the creating (110") of the two-dimensional camera image (21) is accomplished by creating a dark-field camera image of the pallet (3) in the near-infrared frequency spectrum, and wherein the dark-field camera image is recorded from a recording direction (5a) that forms an angle (5b) with the support plane (17a), the angle lying in the range from 10° to 70°, in particular greater than 45°, for example 60°, and extends at an angle in a range from 70° to 110°, in particular that is perpendicular to the main direction of extension.

7. The method according to one of claims 1 to 3, wherein the creation (100) of the contrast-image dataset (23) of the pallet (3) further comprises: - creating (110") the contrast-image dataset (23) by creating a dark-field camera image of the pallet (3) in the near-infrared frequency spectrum from a recording direction (5a) that forms an angle (5b) with the support plane (17a), the angle lying in the range from 10° to 70°, in particular greater than 45°, for example 60°, and extends at an angle in a range from 70° to 110°, in particular that is perpendicular to the main direction of extension.

8. The method according to one of the preceding claims, wherein the creation (100) of the contrast-image dataset (23) of the pallet (3) further comprises: - illuminating (105") the pallet (3) from above with light of a near-infrared frequency spectrum, in particular in the frequency spectrum from 845 nm to 855 nm and / or from 935 nm to 945 nm, from one or more illumination directions (7a).

9. The method according to claim 8, wherein at least one of the illumination directions (7a) forms an angle (7b) with the support plane (17a) which is less than 30° and in particular lies in a range from 10° to 20°, for example approximately 15°, and the at least one of the illumination directions (7a), when projected into the support plane (17a), forms an angle (7c) with a main direction of extension (13a) of the support bar (13), the angle being less than 30° and lying in a range from 10° to 25° in particular, for example approximately 15°.

10. The method according to one of claims 4 to 9, wherein the creating (110') of the two-dimensional camera image (21) or the creating (110") of the two-dimensional dark-field camera image (23) comprises: - recording a plurality of partial camera images, each of which two-dimensionally represent a partial region of the pallet (3), and - putting the partial camera images together to form the two-dimensional camera image (21) or the two-dimensional dark-field camera image (23') of the pallet (3).

11. The method according to one of the preceding claims, wherein the determining (200) of the longitudinal position in the contrast-image dataset (23, 23') comprises: - carrying out a template-matching image analysis using a template (25) formed as an image dataset of a support bar (13) that has a plurality of support bar tips (17), in particular a support bar (13) arranged in an inner region of the pallet (3).

12. The method according to one of the preceding claims, wherein the determining (200) of the longitudinal position in the contrast-image dataset (23, 23') comprises: - totalizing the pixel values of the image pixels of the contrast-image dataset (23, 23') row-by-row corresponding to the main direction of extension (13a), and outputting a first distribution of first pixel-value sums; and - determining a local extreme in the first distribution, and outputting the row of the local extreme as a longitudinal position.

13. The method according to claim 12, further comprising: - totalizing the pixel valves of the image pixels of the contrast-image dataset (23, 23') column-by-column corresponding to the longitudinal direction (13a) in the region of the local extreme, and outputting a second distribution of second pixel-value sums, - determining a plurality of local extremes in the second distribution and outputting the columns of the local extremes and - determining (200) image positions of the regions in the contrast-image dataset (23, 23') that are assigned to the support bar tips (17) on the basis of the corresponding row of the determined local extreme and the columns of the determined local extremes.

14. The method according to one of the preceding claims, wherein the support bars (13) can be suspended at predetermined suspension positions (15) on a longitudinal side (3a) of the pallet (3), and the main direction of extension (13a) extends transversely relative to the longitudinal side (3a) and the pallet (3) has a reference structure (19a) which is identifiable in the contrast-image dataset (23, 23') as a reference pixel arrangement (19b), and wherein the derivation (300) of the suspension position (15) of the support bar (13) in the pallet (3) is accomplished by determining a distance between the reference pixel arrangement (19b) and the longitudinal position transverse to the main direction of extension (13a) in the contrast-image dataset (23, 23') and, on the basis of the distance determined, identifying one of the predetermined suspension positions (15) for the longitudinal position.

15. The method according to one of the preceding claims, wherein the pixel value is a gray value, a color value, or a brightness value.

16. A flatbed machine tool (1), in particular a laser-cutting or plasma-cutting flatbed machine tool, comprising: a pallet (3) with a plurality of multiple support bars (13) suspended at suspension positions (15), wherein each of the support bars (13) has a plurality of support bar tips (17) that define a support plane (17a), a camera (5) configured to create a two-dimensional camera image (21) or a two-dimensional dark-field camera image of the pallet and an evaluation device (11), characterized in that the camera (5) and / or the evaluation device (11) are configured for - creating (100) a two-dimensional contrast-image dataset (23, 23') of the pallet (3), the dataset having a plurality of image pixels, wherein a pixel value and a pixel area unit of the support plane (17a) is assigned to each image pixel, and the contrast-image dataset (23, 23') comprises regions that are assigned to the support bar tips (17) as local pixel-value extremes (23a) in a uniform pixel-value background (23b), and the evaluation device (11) is configured for - determining (200) a longitudinal position in the contrast-image dataset (23, 23') of regions that are assigned to the support bar tips (17) of one of the plurality of support bars (13) using the local pixel-value extremes (23a) and - deriving (300) the suspension position (15) of the support bar (13) in the pallet (3) on the basis of the longitudinal position in the contrast-image dataset (23) and an extent of the pixel area unit in the longitudinal direction (x).

17. The flatbed machine tool (1) according to claim 16, wherein the camera (5) is configured, - to create the two-dimensional camera image (21) of the pallet in the visible or infrared spectrum, wherein the camera image (21) has image pixels arranged in rows and columns, one each of the pixel-area units of the support plane (17a) being assigned to said image pixels, respectively, or - to create a contrast-image dataset (23') by recording the two-dimensional dark-field camera image of the pallet (3) in the near-infrared frequency spectrum from a recording direction (5a).

18. The flatbed machine tool (1) according to claim 17, wherein the evaluation device (11) is configured to carry out a method according to one of the preceding claims and / or comprises a processor (9) which is configured - to Fourier transform the camera image (21) into a frequency domain and to output a transformation-frequency dataset (21a), - to filter the transformation-frequency dataset (21a) in order to emphasize frequencies in the frequency domain that belong to the regions which are assigned to the support bar tips (17), and - to back-transform the filtered transformation-frequency dataset into a using an inverse Fourier transformation and output the back-transformed image dataset as the contrast-image dataset (23).

19. The flatbed machine tool (1) according to one of claims 16 to 18, further comprising at least one illumination device (7) configured to illuminate the pallet (3) with light of a near-infrared frequency spectrum from one illumination directions (7a) and wherein optionally at least one of the illumination directions (7a) forms an angle (7b) with the support plane (17a) which is less than 30° and in particular lies in a range from 10° to 20°, for example approximately 15°, and the at least one of the illumination directions (7a), when projected into the support plane (17a), forms an angle (7c) with a main direction of extension of the support bar, the angle being less than 30° and lying in a range from 10° to 25° in particular, for example approximately 15°.