WO2021148252A1 - Procédé, dispositif et programme informatique de détection d'objet - Google Patents

Procédé, dispositif et programme informatique de détection d'objet Download PDF

Info

Publication number
WO2021148252A1
WO2021148252A1 PCT/EP2021/050193 EP2021050193W WO2021148252A1 WO 2021148252 A1 WO2021148252 A1 WO 2021148252A1 EP 2021050193 W EP2021050193 W EP 2021050193W WO 2021148252 A1 WO2021148252 A1 WO 2021148252A1
Authority
WO
WIPO (PCT)
Prior art keywords
geometric
hash code
features
coordinates
hash
Prior art date
Application number
PCT/EP2021/050193
Other languages
German (de)
English (en)
Inventor
Hendrik Leibrandt
Volker ZÖLLMER
Marc-Oliver BECKER
Original Assignee
Fraunhofer-Gesellschaft zur Förderung der angewandten Forschung e.V.
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Fraunhofer-Gesellschaft zur Förderung der angewandten Forschung e.V. filed Critical Fraunhofer-Gesellschaft zur Förderung der angewandten Forschung e.V.
Publication of WO2021148252A1 publication Critical patent/WO2021148252A1/fr

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/50—Context or environment of the image
    • G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G06F18/231—Hierarchical techniques, i.e. dividing or merging pattern sets so as to obtain a dendrogram

Definitions

  • Embodiments of the present invention deal with a method, a device and a computer program for recognizing an object.
  • a smooth flow of a production line is essential in order, among other things, to achieve a high level of economic efficiency and to meet mandatory safety standards.
  • processes e.g. for manufacturing, processing and checking components
  • processes are linked with one another in such a way that the components can run through the several process steps efficiently and in the correct sequence.
  • the tracking of a component along the production chain is possible, for example, using identification processes or recognition processes, so that the component can be correctly identified and assigned for the further process step.
  • Marking processes that apply an additional marking to the component or emboss the component itself are not only costly due to the additional process steps required for marking, but also change the surface of the component to be identified.
  • Some labeling processes for example, attach transponders to the component or print QR codes on it, which can, however, easily be copied and forged.
  • Recognition methods use, for example, distinctive features that arise on the surface of the object as a result of production, such as the microstructure of printed components.
  • Some detection methods are based on a direct image comparison between a reference component and a component to be determined and therefore require strictly reproducible conditions for reliable detection of the component, such as an exact alignment of the component with respect to a camera.
  • Algorithms such as S SWIFT or SURF generate many false-positive matches for similarly manufactured components or for components with a modified surface, e.g. due to oxidation, and are often not reliable enough.
  • the previous methods for object recognition do not offer an efficient method with a reasonable computing time.
  • the patent US 2013/0071875 A1 relates to a method for identifying biological objects on a substrate.
  • the substrate that carries the biological objects is illuminated with a light beam comprising a specific wavelength range in order to generate an image.
  • the image includes aberrations that are characteristic of the object.
  • An exemplary embodiment of a method for recognizing an object, in particular a component includes registration of geometric features in a recording of the object and generation of a hash code using a hash function on a subset of the geometric features. Furthermore, the method comprises a comparison of the hash code with a reference hash code with regard to a predefined criterion, the reference hash code being obtained using the hash function on a subset of geometric reference features of a reference object. The method also includes generating a transformation rule when the hash code matches the reference hash code with regard to the predefined criterion, the transformation rule mapping the subset of the geometric features of the hash code onto the subset of the geometric reference features of the reference hash code.
  • Geometric features result, for example, from the surface structure of the object.
  • object recognition it is suitable to summarize the geometric features in hash codes so that the method for object recognition can work with a reduced data set. This means that the object can be identified quickly and reliably at the same time, since the spatial relationship between the geometric features is not lost through the merging in hash codes. As a result, it may be sufficient, for example, to compare less data in the form of hashes and reference hashes with one another without the quality of the method being impaired.
  • a suitably selected predefined criterion can determine whether a similarity between the hash code and the reference hash code is sufficient to generate a transformation rule.
  • the transformation rule can be used to reliably check whether the object actually matches a reference object. By applying the transformation rule, all geometric features can be used for a comparison between see the object and the reference object are used to reliably identify the object.
  • the hash function generates a hash code based at least on the two-dimensional coordinates of a first geometric feature xl, yl, a second geometric feature x2, y2 and a third geometric feature x3, y3.
  • the coordinates of the third geometric feature lie within a circle with a diameter determined by the distance between the coordinates of the first and second geometric features.
  • a local coordinate system is defined so that the generated hash code includes the spatial relationship of the geometric features.
  • the amount of data is reduced from a total of three two-dimensional vectors of the three geometric features to one two-dimensional vector of the hash code.
  • the hash code becomes four-dimensional.
  • a four-dimensional hash code is more characteristic than a two-dimensional hash code.
  • the probability that a four-dimensional hash code and a reference hash code incorrectly coincidentally coincides is lower. On the one hand, this can increase the reliability of the method for object recognition. On the other hand, the computing power can be reduced, since fewer false-positive matches have to be discarded after the transformation rule has been applied.
  • the predefined criterion is fulfilled when the hash code is at a predefined distance from the reference hash code.
  • the predefined distance such as the Euclidean distance, allows a hash code match with a defined fuzziness.
  • the search radius for a similar reference hash code can be restricted by the predefined distance.
  • the method further comprises a search for the reference hash code in an index structure with a multiplicity of stored possible reference hash codes.
  • An efficient algorithm for a higher-dimensional distance search of reference hash codes can be carried out on the basis of the index structure.
  • the search scope can be reduced by the index structure of the reference database, since not all reference hash codes stored and indexed in it have to be compared with the hash code.
  • the method further comprises transforming all geometric features of the object with the transformation rule in order to obtain transformed geometric features.
  • the method further includes checking the transformed geometric features and the geometric reference features for geometric correspondence, with the object being identified as belonging to the class of the reference object if a correspondence is established. For a final, reliable assessment of whether the object corresponds to the reference object, all geometric features can be taken into account through the mapping using the transformation rule and checked for conformity.
  • the method further comprises an adaptation of the transformation rule in order to improve a mapping of the subset of the geometric features of the hash code onto the subset of the geometric reference features of the reference hash code.
  • the transformation rule can also be adapted for fine adjustment so that the alignment of the geometric feature pairs can be improved.
  • the method furthermore comprises a removal of periodic structures in the receptacle of the object, which were generated as artifacts from a manufacturing method of the object. Removing artifacts can reduce the number of false-positive matches.
  • the method can deliver faster results, since fewer incorrectly assumed matches between hash code and reference hash code are discarded.
  • the method further comprises generating the recording of the object with directional illumination of the object with an angle of incidence of less than 30 ° to the surface of the object. This allows high-contrast images of the object to be recorded, as geometric features such as depressions can be more clearly identified.
  • An embodiment of a device for recording an object comprises a camera for taking a picture of a top view of the object and a light source for directed illumination of the object with an angle of incidence of less than 30 ° to the surface of the object. This allows high-contrast images of the object to be recorded, as geometric features such as depressions can be more clearly identified.
  • the device further comprises an optical component which is designed to absorb scattered light, the optical component being arranged between the camera and the object.
  • an optical component which is designed to absorb scattered light, the optical component being arranged between the camera and the object.
  • An embodiment of a device for recognizing an object comprises an analysis circuit which is designed to register geometric features in a recording of the object and to generate a hash code using a hash function on a subset of the geometric features. Furthermore, the analysis circuit is designed to compare the hash code with a reference hash code with regard to a predefined criterion, the reference hash code being obtained using the hash function on a subset of geometric reference features of a reference object. Furthermore, the analysis circuit is designed det to generate a transformation rule when the hash code matches the reference hash code with regard to the predefined criterion, the transformation rule mapping the subset of the geometric features of the hash code onto the subset of the geometric reference features of the reference hash code.
  • Geometric features result, for example, from the surface structure of the object.
  • object recognition it is suitable to summarize the geometric features in hash codes so that the method for object recognition can work with a reduced data set. This means that the object can be identified quickly and reliably at the same time, since the spatial relationship between the geometric features is not lost through the merging in hash codes. As a result, it may be sufficient, for example, to compare less data in the form of hashes and reference hashes with one another without the quality of the method being impaired.
  • a suitably selected predefined criterion can determine whether a similarity between the hash code and the reference hash code is sufficient to generate a transformation rule.
  • the transformation rule can be used to reliably check whether the object actually matches a reference object. By applying the transformation rule, all geometric features can be used for a comparison between the object and the reference object in order to reliably identify the object.
  • the analysis circuit is further designed so that the hash function generates a hash code at least based on the two-dimensional coordinates of a first geometric feature xl, yl, a second geometric feature x2, y2 and a third geometric feature x3, y3.
  • the coordinates of the third geometric feature lie within a circle with a diameter determined by the distance between the coordinates of the first and second geometric features.
  • the coordinates x3, y3 of the third geometric feature represented in the local two-dimensional coordinate system defined by the coordinates of the first geometric feature xl, yl as the origin 0, 0 of the coordinate system and defined by the coordinates of the second geometric feature x2, y2 1, 1 is spanned, form the hash code [x3 ', y3'].
  • a local coordinate system is defined so that the generated hash code includes the spatial relationship of the geometric features.
  • the data amount of a total of three two-dimensional vectors of the three geometric features on one two-dimensional vector of the hash code is used to form the hash code [x3 ', y3'].
  • the hash code becomes four-dimensional.
  • a four-dimensional hash code is more characteristic than a two-dimensional hash code.
  • the probability that a four-dimensional hash code and a reference hash code incorrectly coincidentally coincides is lower. On the one hand, this can increase the reliability of the method for object recognition. On the other hand, the computing power can be reduced, since fewer false-positive matches have to be discarded after the transformation rule has been applied.
  • the predefined criterion is met when the hash code is at a predefined distance from the reference hash code.
  • the predefined distance allows a hash code match with a defined fuzziness.
  • the search radius for a similar reference hash code can be restricted by the predefined distance.
  • the analysis circuit is also designed to search for the reference hash code in an index structure with a plurality of stored possible reference hash codes.
  • An efficient algorithm for a higher-dimensional distance search of reference hash codes can be carried out on the basis of the index structure.
  • the search scope can be reduced by the index structure of the reference database, since not all reference hash codes stored and indexed in it have to be compared with the hash code.
  • the analysis circuit is further designed to transform all geometric features of the object with the transformation rule in order to obtain the transformed geometric features and to check the transformed geometric features and the geometric reference features for geometric correspondence, with the object being determined is identified as belonging to the class of the reference object. For a final, reliable assessment of whether the object corresponds to the reference object, all geometric features can be taken into account through the mapping using the transformation rule and checked for conformity.
  • the analysis circuit is also designed to adapt the transformation rule in order to improve a mapping of the subset of the geometric features of the hash code onto the subset of the geometric reference features of the reference hash code.
  • the transformation rule can also be adapted so that the alignment of the geometric feature pairs can be improved.
  • the analysis circuit is also designed to remove periodic structures in the receptacle of the object that were generated as artifacts from a manufacturing process for the object. Removing periodic structures can reduce the number of false-positive matches. In addition, the method can deliver faster results, since fewer incorrectly assumed matches between hash code and reference hash code are discarded.
  • An exemplary embodiment of the present invention comprises a computer program with program code which executes a method according to the preceding description using a programmable processor.
  • any compatible device can be upgraded to recognize an object using the method described.
  • 1 shows an exemplary embodiment of a device for recognizing an object, in particular components
  • 2 shows an exemplary embodiment for generating a hash code using a hash function on a subset of geometric features
  • FIG. 3 shows an exemplary embodiment of geometric features in a recording of an object
  • FIG. 4 shows an exemplary embodiment of a database with four-dimensional hash codes from five objects
  • FIG. 5 shows an exemplary embodiment of a device for receiving an object, in particular a component
  • FIG. 6 shows a flow diagram of an exemplary embodiment for a method for recognizing an object, in particular components.
  • the device 100 comprises an analysis circuit 101 which is designed to register geometric features in a recording of the object 102.
  • the recording of the object 102 can be a two-dimensional image of a printed component that was recorded with a camera, for example.
  • the recording can be made immediately before the start of the method, or existing recordings can be used for the method that are obtained from a storage medium via a network, for example.
  • the geometric features can be characterized by distinctive areas in the image, such as mountains and / or valleys of a superficial microstructure of the object 102.
  • the object 102 should be recorded in a suitable manner so that the geometric features can be represented in the best possible and reproducible manner.
  • Geometric features are, for example, two-dimensional pixel coordinates of local maxima in the image of component 102.
  • Each set consisting of at least three different geometric features can be converted into a hash code using a hash function. A smaller amount of data can thus be used for the further method for object recognition, since three or more spatial coordinates of the geometric features are combined in the multidimensional hash code.
  • the method compares a first hash code with a reference hash code with regard to a predefined criterion.
  • the reference hash code is generated by applying the hash function to a subset generated by geometric reference features of a reference object and can for example be stored in a database.
  • the predefined criterion can be met if the hash code is at a predefined distance from the reference hash code. In this case, a search is made for similar reference hash codes in an index structure with a large number of stored possible reference hash codes, for example using a suitable algorithm. If a hash code matches a reference hash code, a transformation rule is generated, the transformation rule mapping the geometric features that contribute to the hash code onto the geometric reference features that contribute to the reference hash code.
  • the transformation rule can also be adapted for better mapping.
  • the transformation rule can then be applied to all registered geometric features of the object in order to obtain transformed geometric features. If the transformed geometric features match the geometric reference features, the object is identified as the reference object. If there is no geometric match, a next hash code based on other geometric features of the object 102 can be used. The object 102 is deemed not to have been identified if there is no geometric correspondence between the transformed geometric features and the geometric reference features for all of the hash codes considered.
  • the method can thus identify the object 102 with a hash code that matches a reference hash code. This can be of particular advantage if only part of the object 102 was recorded in the image recording, e.g. because the component 102 was partially changed or damaged.
  • the method can provide reliable results for object recognition, since in the final evaluation for the recognition of the object 102 all transformed geometric features can be compared with the geometric reference features.
  • FIG. 2 shows the generation of a hash code 210 using a hash function on a subset of the geometric features 211-214 according to an exemplary embodiment.
  • the hash code 210 which is identified by the letter H, can be represented in this exemplary embodiment by a four-dimensional vector.
  • the four geometric features 211-214 which are used to generate a hash code in accordance with the illustrated hash function are identified in FIG. 2 by the letters AD and each correspond to two-dimensional coordinates in a defined coordinate system.
  • the coordinate system can be selected as desired or, for example, be based on the coordinate system of a device for recording the object, for example a light microscope.
  • a first geometric feature 211 is identified by the letter A and a second geometric feature 212 is identified by the letter B.
  • the first geometric feature 211 and the second geometric feature 212 are suitable to be selected from the subset of the geometric features 211-214 in such a way that the distance between the first geometric feature 211 and the second geometric feature 212 is greatest.
  • the first geometric feature 211 and the second geometric feature 212 can span a circle 215 which has the distance between the first geometric feature 211 and the second geometric feature 212 as a diameter.
  • the first and second geometric features 211 and 212 can span a local coordinate system 216, the origin 0, 0 of the local coordinate system 216 being defined by the first geometric feature 211 and the coordinate 1,1 of the local coordinate system 216 by the second geometric feature 212 .
  • the third and fourth geometric features 213 and 214 one can consider coordinates that lie within the spanned circle 215.
  • the coordinates x3, y3 of the third geometric feature 213 and the coordinates x4, y4 of the fourth geometric feature 214 can be represented in the local coordinate system 216 by the coordinates x3 ', y3' and x4 ', y4'.
  • the four-dimensional hash code 210 can be formed by the coordinates [x3 ', y3', x4 ', y4'].
  • the hash codes 210 of an object are thus made up of differing subsets formed by geometric features.
  • the hash code 210 describes a local pattern of the object in a single coordinate and is invariant to translation, rotation and scaling of the geometric features 211-214.
  • mapping by the hash function is tolerant with regard to small disturbances. For example, small changes in the positions of the geometric features 211-214 become noticeable through small changes in the coordinates of the hash code 210. This means that the method can also work with small surface changes of the component and can represent a significant improvement with regard to other detection methods.
  • the exemplary embodiment for generating a hash code 210 in FIG. 2 comprised four geometric features 211-214.
  • the hash code 210 can alternatively be formed from at least three geometric features.
  • the definition of the local coordinate system can take place according to the same conditions as in the exemplary embodiment mentioned in FIG. 2.
  • Hash codes of a higher dimensional order are more characteristic, so that a false, random match between the hash code and a reference hash code is less likely. This can reduce the computing time for recognizing the object, since fewer false-positive matches are generated and discarded.
  • hash functions can be any, the spatial relationship between the geometric features being recorded in the generated hash code and the hash code being invariant to translation, rotation and scaling of the geometric features.
  • geometric features in the recording of the object can take place on the basis of a defined criterion or an algorithm.
  • geometric features can be coordinates of pixels, the pixels being characterized, for example, by maximum or minimum intensities, edges, colors, contrasts, absolute or relative deviations from a target value, threshold values, distances and much more.
  • pixel coordinates can be determined by parameters, the parameters resulting from a compensation calculation, such as, for example, fitting a Gaussian curve to measurement data.
  • Geometric features can also be determined, for example, by feature extraction algorithms such as SIFT, SURF, BRISK, ORB, SWIFT, or by machine vision techniques and many others. FIG.
  • FIG. 3 shows an exemplary embodiment for a set of geometric features in a receptacle 320 of an object.
  • the geometric features are numbered consecutively with the letter “P” and shown in a two-dimensional coordinate system.
  • groups of four neighboring geometric features can be considered.
  • adjacent geometric features such as 311-314, for example, are connected to one another by edges.
  • the exemplary embodiment shows that a minimum distance can be established so that geometric features are declared as neighbors.
  • the geometric feature 315 is not a neighbor of the geometric feature 313, since the coordinates of the two features do not have the specified minimum distance.
  • Hash codes can be generated from the set of four different, neighboring, geometric features.
  • permutations can be considered so that other or new neighbors can be assigned to the geometric features.
  • new subsets consisting of neighboring geometric features can be formed, whereby the subsets considered should differ in order to avoid symmetrical hash codes.
  • variations in the combinations of neighbors of neighbors can also be taken into account.
  • the search radius for hash formation can be extended over two edges. This can, for example, prevent network connectivity from being interrupted due to the lack of individual points. In this way, the method can be improved, for example with regard to its robustness.
  • Different methods can be used to determine the subsets of geometric hash codes. The method can restrict the subset of geometric features on the basis of criteria, such as a defined minimum distance and / or maximum distance.
  • the coordinate system with the coordinates of the geometric features can be subdivided into grids, so that subsets of geometric features can be viewed within a grid cell.
  • further examples include any method that can subdivide the registered geometric features of the object into subsets using a predefined method so that hashing can be applied.
  • FIG. 4 shows an embodiment of 384 hash codes in a database.
  • four-dimensional hash codes were generated from the subsets of 50 geometric features of five objects each.
  • the four coordinates X, Y, Z, and W of the hash codes are plotted and compared on each axis.
  • the hash codes in the exemplary embodiment in FIG. 4 can be viewed as reference hash codes in a database, the reference hash codes being obtained using the hash function on a subset of geometric reference features of the reference objects.
  • the hash codes of the object are compared with the reference hash codes of the reference objects with regard to a predefined criterion.
  • the predefined criterion can determine whether the hash code under consideration is similar to the reference hash code in the database, i.e. e.g. that the coordinates between the hash code and the reference hash code differ slightly.
  • the predefined criterion is met when the hash code is at a predefined distance from the reference hash code.
  • the predefined distance can, for example, be an absolute value which, for example, considers the Euclidean distance based on the hash code. All reference hash codes within this distance can then meet the predefined criterion, so that the hash code can be considered to match the reference hash code.
  • the reference hash code can be searched for in an index structure with a multiplicity of stored possible reference hash codes. So that the four-dimensional distance search can run efficiently, it is suitable to create a database that manages a hierarchical search index, for example in the form of a KD tree, an area tree or a ball tree.
  • search methods can be any, such as in lists, trees, graphs, and include manual or automatic methods that efficiently compare hash codes with reference hash codes with regard to a predefined criterion. If the predefined criterion is met, the hash code and reference hash code can be regarded as matching.
  • a transformation rule is generated, the transformation rule mapping the subset of the geometric features of the hash code onto the subset of the geometric reference features of the reference hash code.
  • the transformation rule can, for example, be a function or matrix which assigns the geometric features of the hash code under consideration to the geometric reference features of the associated reference hash code. Using this transformation rule, all geometric features of the object can be mapped onto transformed geometric features.
  • the transformation rule can be adapted in order to improve a mapping of the subset of the geometric features of the hash code onto the subset of the geometric reference features of the reference hash code.
  • a RANSAC algorithm can be used to compute a fine alignment between the geometric features and the reference geometric features and to identify outliers. Further methods for adapting the transformation rule include, for example, compensation methods such as the method of least squares.
  • the transformed geometric features and the geometric reference features are checked for geometric correspondence.
  • the geometric correspondence can be made using algorithms such as RANSAC, or using machine vision methods, or using a specified criterion.
  • a score can be calculated which shows the number of exact matches of the pairs of features, the number of transformed geometric features without a matching geometric reference feature, the number of geometric reference features without a matching transformed geometric feature and the homography as a reference point for overlapping image areas taken into account.
  • a simplified example of the score is, for example, the ratio of the number of exact matches versus the number of total pairs of features between transformed geometric features and reference features. For example, areas of great agreement can also be assessed, since damage changes not just individual features, but entire areas. If the score exceeds a threshold value, for example, the object can be identified as belonging to the class of the reference object. Otherwise, the object is considered unidentified, so that the next hash code can be used.
  • FIG. 5 shows an exemplary embodiment of a device 500 for receiving an object 501, in particular a component.
  • the device comprises a camera 502 for taking a picture of a top view of the object 501 and a light source 503 for the directed illumination of the object 501 with an angle of incidence smaller than 30 ° to the surface of the object.
  • the device 500 can be used, for example, to appropriately detect and display the structure of the surface of the object 501.
  • One or more special light sources 503 can be used to highlight the surface topology.
  • the light source can, for example, be arranged in a circle on the device 500 so that, for example, the light can be incident on the center of the observed object surface via a light guide at a very flat angle.
  • the directional illumination of the object surface should take place in such a way that the light emitted from the light source 503 can be as parallel as possible to the object surface, for example at angles of 1.0 °, 1.5 °, 2.0 °, etc.
  • the light source 503 can be a Any object that emits electromagnetic radiation in the visible range so that the object can be illuminated directly.
  • the device 500 further comprises an optical component 504 which is designed to absorb scattered light, the optical component 504 being arranged between the camera 502 and the object 501.
  • the optical component 504 can be used to absorb scattered light that is produced, for example, by reflection on the surface of the object 501 or the device 500.
  • the optical component 504 can be, for example, a screen, a filter or any other component that can absorb scattered light due to its material properties.
  • the optical component 504 can be used to create recordings of the object 501 with a good contrast ratio, so that geometric features of the object 501 can be clearly and reproducibly registered and recognized.
  • the camera 502 of the device 500 can take a picture of the top view of the object 501.
  • cameras are digital cameras, compact cameras, single-lens reflex cameras, system cameras, or any type of camera that is implemented as part of a system, such as smartphones.
  • microscopic methods can also be used for image recording, in particular dark field microscopy.
  • methods for preprocessing the recordings of the objects can be carried out.
  • a number of filters can be applied to the image in order to remove unwanted parts of the image and to reinforce characteristic geometric features.
  • Further exemplary embodiments include the removal of periodic structures in the recording of the object, which were generated as artifacts from a manufacturing process for the object. In the printing process in particular, periodic disturbances can occur in the component.
  • the periodic artifacts are not object-specific and are therefore not suitable as a geometric feature for object recognition.
  • the method for object recognition can be improved, since the artifacts are not registered as a geometric feature.
  • the frequency domain of the interfering, periodic structures of the image can be filtered by means of a spectral analysis. By removing periodic structures, the statistical distribution of the geometric features can change significantly, so that the method for object recognition can be significantly improved.
  • FIG. 6 shows a flowchart of an exemplary embodiment of a method for recognizing an object, in particular a component 700.
  • the method comprises a registration of geometric features in a recording of the object 701 and generation of a hash code using a hash function on a subset of the geometric features 702 Furthermore, the method includes a comparison of the hash code with a reference hash code with regard to a predefined criterion, the reference hash code being obtained 703 using the hash function on a subset of geometric reference features of a reference object.
  • the method also includes generating a transformation rule if the hash code matches the reference hash code with regard to the predefined criterion, the transformation rule applying the subset of the geometric features of the hash code to the subset of the geometric reference features of the reference hash code it maps 704.
  • the developed invention can operate in the optical field and includes, for example, a camera that records image information in the visible spectrum.
  • the identification of the object can be based on the uniqueness of a number of surface features which, taken on their own, can be ambiguous, but become unmistakable as a whole. Characteristics that are as concise as possible can be used, which remain recognizable even with limited changes to the surface, such as due to aging, wear and tear, soiling, etc.
  • the surface of the component can be mapped. Distinctive features can be used for this. These should be clearly recognizable, stable, but also available in sufficient numbers. This map does not describe the characteristics themselves, because they show little variance, but their spatial distribution and relationship to one another. In this way, it is possible to reconstruct this assignment with just a section or an incomplete recording. Geometric hashing can be used for the method.
  • Geometric hash codes can represent an efficient method for matching geometric features and can solve the NP-heavy point set registration problem via the use of an index structure in order to reduce the search effort considerably. This shifts complexity into the creation of the index and allows search queries with constant time complexity. The basic requirement is a statistical distribution of the geometric features, which also results in the need for preprocessing to remove any periodic structures that may be present.
  • a hashing method maps a group of four neighboring geometric features into a four-dimensional hash code.
  • the geometric hash codes can have some desirable properties. These are: is invariant to rotation and scaling, tolerates low perspective distortions, is robust to position noise, so that a small change in position of the geometric features causes a small change in the hash code. Furthermore, the method can also work with highly noisy geometric features with a large number of outliers from the query and reference set, for example caused by damage or changes to the surface, because of only partially matching image sections Obscuration or changed camera position. Furthermore, the method can have very good scalability, so that millions of objects can be managed without any problems without affecting the search time. Furthermore, the method can have a low probability of collision, ie there is a low risk of mix-ups due to the high dimensionality of the hash codes used and good utilization of the key space.
  • One advantage of the invention can be a reliable and fast identification method for printed components that does not require any additional component marking.
  • the cost-neutrality in production can enable the tracking of even the smallest and cheapest components in both large and small quantities.
  • the advantage over other optical methods can be the high speed of detection and the good scalability, the robustness against variable recording conditions, as well as changes and / or damage to the component.
  • the same image section does not necessarily have to be compared as long as the overlap is sufficient.
  • a possible rotation does not affect the process at all. Due to its robustness, there are no high requirements for image acquisition, and handy devices are also conceivable. Because only the position, but not the exact structure of a feature, is included, the method can be insensitive to a change in the same. Other methods tolerate large areas of coverage or damage to the surface, but require areas for identification that are unchanged. Minor changes affecting the entire surface, e.g. due to aging or corrosion, lead to a significant decrease in detection accuracy.
  • Technical fields of application of the invention include, for example, the tracking of components in production and assembly, in areas where conventional marking systems cannot be used, for example due to a lack of space, impairment of the Component function, problems with the application or for cost reasons.
  • the method is also suitable for differentiating between an original part and a forgery, for example in the area of counterfeit protection.
  • Examples can furthermore be a computer program with a program code for executing one or more of the above methods or refer to them when the computer program is executed on a computer or processor. Steps, operations or processes of various methods described above can be carried out by programmed computers or processors. Examples can also include program storage devices, e.g. B. digital data storage media that are machine, processor or computer readable and encode machine, processor or computer executable programs of instructions. The instructions perform or cause some or all of the steps in the methods described above.
  • the program storage devices may e.g. B. digital storage, magnetic storage media such as magnetic disks and tapes, hard disk drives or optically readable digital data storage media or be.
  • FIG. 1 For purposes of this specification
  • a function block referred to as “means for ...” executing a specific function can relate to a circuit which is designed to execute a specific function.
  • a “means for something” can be implemented as a “means designed for or suitable for something”, e.g. B. a component or a circuit designed for or suitable for the respective task.
  • each function block designated as “means”, “means for providing a signal”, “means for generating a signal”, etc. may be in the form of dedicated hardware, e.g. B “a signal provider”, “a signal processing unit”, “a processor”, “a controller” etc. as well as being implemented as hardware capable of executing software in conjunction with the associated software.
  • the functions can be provided by a single dedicated processor, by a single shared processor, or by a plurality of individual processors, some or all of which can be shared.
  • DSP digital signal processor hardware
  • ASIC application-specific integrated circuit
  • FPGA Field Programmable Gate Array
  • ROM Read Only Memory
  • RAM Random Access Memory
  • non-volatile storage device storage.
  • Other hardware conventional and / or custom, can also be included.
  • a block diagram may represent a high level circuit diagram that implements the principles of the disclosure.
  • a flowchart, sequence diagram, state transition diagram, pseudocode, and the like may represent various processes, operations, or steps, for example, essentially represented in computer-readable medium and thus performed by a computer or processor, whether or not such Computer or processor is shown explicitly.
  • Methods disclosed in the description or in the claims can be implemented by a device having a means for performing each of the respective steps of these methods. It should be understood that the disclosure of multiple steps, processes, operations, or functions disclosed in the specification or claims should not be construed as being in order, unless explicitly or implicitly otherwise, e.g. B. is given for technical reasons.
  • a single step, function, process, or operation may include and / or be broken into multiple sub-steps, functions, processes, or operations. Such partial steps can be included and part of the disclosure of this individual step, unless they are explicitly excluded.
  • each claim can stand on its own as a separate example. While each claim may stand on its own as a separate example, it should be noted that although a dependent claim in the claims may refer to a particular combination with one or more other claims, other examples also combine the dependent claim with the subject matter of each other dependent or independent claims. Such combinations are explicitly suggested here, unless it is stated that a specific combination is not intended. It is also intended to include features of a claim for any other independent claim, even if that claim is not made directly dependent on the independent claim.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Multimedia (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Image Analysis (AREA)

Abstract

Un mode de réalisation concerne un procédé de détection d'un objet, en particulier un composant 700, comprenant les étapes consistant à enregistrer des caractéristiques géométriques dans un enregistrement de l'objet 701 et à générer un code de hachage, appliquant ainsi une fonction de hachage à un sous-ensemble des caractéristiques géométriques 702. Le procédé comprend en outre l'étape consistant à comparer le code de hachage à un code de hachage de référence par rapport à un critère prédéfini, le code de hachage de référence étant produit par application de la fonction de hachage à un sous-ensemble de caractéristiques de référence géométriques d'un objet de référence, et une recherche étant effectuée pour le code de hachage de référence dans une structure d'index hiérarchique avec une pluralité de codes de hachage de référence possibles stockés. Le procédé comprend en outre l'étape consistant à générer une règle de transformation en cas de correspondance entre le code de hachage et le code de hachage de référence par rapport au critère prédéfini, ladite règle de transformation mettant en correspondance le sous-ensemble de caractéristiques géométriques du code de hachage avec le sous-ensemble de caractéristiques de référence géométriques du code de hachage de référence 704.
PCT/EP2021/050193 2020-01-22 2021-01-07 Procédé, dispositif et programme informatique de détection d'objet WO2021148252A1 (fr)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
DE102020101443.0 2020-01-22
DE102020101443.0A DE102020101443A1 (de) 2020-01-22 2020-01-22 Ein Verfahren, eine Vorrichtung, ein Computerprogramm zur Erkennung eines Objekts und eine Vorrichtung zur Aufnahme eines Objekts

Publications (1)

Publication Number Publication Date
WO2021148252A1 true WO2021148252A1 (fr) 2021-07-29

Family

ID=74183144

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/EP2021/050193 WO2021148252A1 (fr) 2020-01-22 2021-01-07 Procédé, dispositif et programme informatique de détection d'objet

Country Status (2)

Country Link
DE (1) DE102020101443A1 (fr)
WO (1) WO2021148252A1 (fr)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116861023A (zh) * 2023-09-01 2023-10-10 北京达美盛软件股份有限公司 一种在三维空间中确定相同几何对象的方法及系统

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20020181780A1 (en) * 2001-06-05 2002-12-05 Matrox Electronic Systems Ltd. Geometic hashing method for model-based recognition of an object
US20130071875A1 (en) 2010-06-01 2013-03-21 Vbact Ltd. Method and system for the detections of biological objects
US9177224B1 (en) * 2013-03-14 2015-11-03 Amazon Technologies, Inc. Object recognition and tracking

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20020181780A1 (en) * 2001-06-05 2002-12-05 Matrox Electronic Systems Ltd. Geometic hashing method for model-based recognition of an object
US20060126943A1 (en) 2001-06-05 2006-06-15 Christian Simon Geometric hashing method for model-based recognition of an object
US20130071875A1 (en) 2010-06-01 2013-03-21 Vbact Ltd. Method and system for the detections of biological objects
US9177224B1 (en) * 2013-03-14 2015-11-03 Amazon Technologies, Inc. Object recognition and tracking

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
"Geometric hashing", WIKIPEDIA-EINTRAG, 10 December 2020 (2020-12-10)
ALEXANDRE R J FRANÃ OIS ET AL: "Generic shape learning and recognition", 13 April 1996, OBJECT REPRESENTATION IN COMPUTER VISION II, SPRINGER BERLIN HEIDELBERG, BERLIN, HEIDELBERG, PAGE(S) 287 - 320, ISBN: 978-3-540-61750-1, XP019187453 *
HAIM J. WOLFSON: "Geometric Hashing: An Overview", IEEE COMPUTATIONAL SCIENCE AND ENGINEERING, vol. 4, no. 4, 1997, pages 10 - 21, XP001120599, DOI: 10.1109/99.641604

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116861023A (zh) * 2023-09-01 2023-10-10 北京达美盛软件股份有限公司 一种在三维空间中确定相同几何对象的方法及系统
CN116861023B (zh) * 2023-09-01 2023-12-12 北京达美盛软件股份有限公司 一种在三维空间中确定相同几何对象的方法及系统

Also Published As

Publication number Publication date
DE102020101443A1 (de) 2021-07-22

Similar Documents

Publication Publication Date Title
DE60210348T2 (de) Bildidentifizierungssystem
EP2417561B1 (fr) Code a deux dimensions et procede
EP0780002B1 (fr) Procede et appareil pour reconstruir de structures constituees de lignes sous forme d'une trame
EP1395945B1 (fr) Procede de detection de falsification dans la reconnaissance d'empreintes digitales au moyen d'une classification de texture d'images a valeurs de gris differentielles
DE102007035884B4 (de) Linienrauschunterdrückungsvorrichtung, -verfahren und -programm
DE60109278T2 (de) Verfahren und Gerät zur Lokalisierung von Schriftzeichen in Bildern aus einer Digitalkamera
DE102006059663A1 (de) Vorrichtung, Verfahren und Computerprogramm zum Identifizieren eines Verkehrszeichens in einem Bild
DE112020005864T5 (de) Verfahren und Einrichtung zur Verifizierung der Authentizität eines Produkts
EP1897067B1 (fr) Procede et dispositif de detection d'une monnaie par utilisation de son image estampee
WO2021148252A1 (fr) Procédé, dispositif et programme informatique de détection d'objet
WO2008034599A2 (fr) Procédé et dispositif de traitement d'image
DE102010037448A1 (de) Verfahren und Vorrichtung zur Erfassung von Leergutbehältern
DE102005049017A1 (de) Verfahren zur Segmentierung in einem n-dimensionalen Merkmalsraum und Verfahren zur Klassifikation auf Grundlage von geometrischen Eigenschaften segmentierter Objekte in einem n-dimensionalen Datenraum
DE202020103679U1 (de) Vorrichtung zur Lage- und Positionserkennung von Markierungen und Computerprogrammprodukt
DE102009016598A1 (de) Zweidimensionaler Symbolcode und Verfahren zum Lesen des Symbolcodes
DE102004028372A1 (de) Verfahren und Vorrichtung zur Segmentierung einer digitalen Abbildung von Zellen
EP2859529A1 (fr) Codage de blobs
DE10118485B4 (de) Automatische Identifikation und/oder Verifikation linienhafter Texturen wie Fingerabdrücke
EP3142068B1 (fr) Procede de detection tridimensionnelle d'objets
WO2003005290A2 (fr) Procede d'estimation qualitative de materiaux
WO2018007619A1 (fr) Procédé et dispositif pour catégoriser une surface de rupture d'un élément
DE112016007447T5 (de) Verfahren zum Vergleichen einer ersten Zeichnung mit einer zweiten Zeichnung durch Vergleichen ihrer Zeichnungselemente
EP2394250A1 (fr) Procédé et dispositif pour vérifier des documents par utilisation d'une transformation en ondelettes
AT408377B (de) Verfahren und vorrichtung zur prüfung bzw. untersuchung von gegenständen
EP2407910B1 (fr) Procédé et dispositif de création d'une base de données d'objets et de recherche d'objets

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21700267

Country of ref document: EP

Kind code of ref document: A1

122 Ep: pct application non-entry in european phase

Ref document number: 21700267

Country of ref document: EP

Kind code of ref document: A1