Image inspection device
Patent Information
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- ASTEMO LTD
- Filing Date
- 2026-01-15
- Publication Date
- 2026-08-06
Smart Images
Figure JP2026001077_06082026_PF_FP_ABST
Abstract
Description
Imaging inspection apparatus
[0001] The present invention relates to an imaging inspection apparatus.
[0002] Patent Document 1 discloses an imaging inspection apparatus that geometrically calculates a captured image of the cylindrical side surface of a test object from three-dimensional data of the cylindrical side surface, corrects it to an image uniformly scaled and developed, and recombines the corrected image.
[0003] Japanese Unexamined Patent Application Publication No. 2009 - 128261
[0004] In Patent Document 1, for a test object having a three-dimensional curved surface, even if there is three-dimensional data, an image with uniform scale cannot be generated, so the appearance inspection of a test object having a three-dimensional curved surface cannot be performed. One of the objects of the present invention is to provide an imaging inspection apparatus that can accurately measure the size of a defect on the three-dimensional curved surface of a test object having a three-dimensional curved surface and prevent overlooking of the defect.
[0005] An imaging inspection apparatus for inspecting a defect in an inspection area of a test object having a three-dimensional curved surface according to an embodiment of the present invention includes an image storage unit that acquires and stores a captured image obtained by imaging an inspection area of the test object with an imaging device, a coordinate information storage unit that stores coordinate information of the distance between the imaging device and the inspection area, a test object shape information storage unit that stores shape information of the test object as test object shape information, an image extraction unit that extracts a defect candidate area image from the captured image, a front-facing viewpoint calculation unit that calculates a front-facing viewpoint facing the defect candidate area image using the coordinate information and the test object shape information, an image viewpoint conversion unit that converts the defect candidate area image into a defect candidate area front-facing image viewed from the front-facing viewpoint, and a defect determination unit that determines whether a defect candidate in the defect candidate area front-facing image is a defect.
[0006] According to an embodiment of the present invention, the size of a defect on the three-dimensional curved surface of a test object having a three-dimensional curved surface can be accurately measured, and overlooking of the defect can be prevented.
[0007] This is an overall schematic diagram of the image inspection apparatus 1 of Embodiment 1. This is the main flowchart showing the image inspection flow in Embodiment 1. This is a time chart showing the direct viewpoint calculation of the direct viewpoint calculation unit in Embodiment 1. This is a time chart showing the defect determination process flow in Embodiment 1. This is a time chart showing the direct viewpoint calculation of the direct viewpoint calculation unit in Embodiment 2. This is an overall schematic diagram of the image inspection apparatus 1 of Embodiment 3. This is an overall schematic diagram of the image inspection apparatus 1 of Embodiment 4. This is an overall schematic diagram of the image inspection apparatus 1 of Embodiment 5. This is a time chart showing the defect determination process flow in Embodiment 6.
[0008] [Embodiment 1] Figure 1 is an overall schematic diagram of the image inspection apparatus 1 of Embodiment 1.
[0009] The image inspection apparatus 1 of Embodiment 1 comprises a camera (imaging device) 2 with a photographic lens with low lens distortion and a computer 3. The camera 2 images a propeller (object to be inspected) 17 having a three-dimensional curved surface. The computer 3 is, for example, a personal computer and comprises memory 4 and a CPU 5. The memory 4 comprises an image storage unit 6, a coordinate information storage unit 7, an object to be inspected shape information storage unit 8, a defect candidate area image storage unit 9, a frontal viewpoint storage unit 10, a frontal image storage unit 11, and a judgment result storage unit 12. The image storage unit 6 stores one image of the inspection area of the propeller 17 captured in one shot by the camera 2. The coordinate information storage unit 7 stores coordinate information of the distance between the camera 2 and the inspection area of the propeller 17. The object to be inspected shape information storage unit 8 stores object to be inspected shape information of the propeller 17, which has been prepared in advance (design dimensions, shape data, etc.). The defect candidate region image storage unit 9 stores the defect candidate region image extracted from the captured image by the image extraction unit 13, which will be described later. The frontal viewpoint storage unit 10 stores the frontal viewpoint that is directly facing the defect candidate region image calculated by the frontal viewpoint calculation unit 14, which will be described later. The frontal image storage unit 11 stores the defect candidate region frontal image, which is converted from the defect candidate region image to an image viewed from the frontal viewpoint by the image viewpoint conversion unit 15, which will be described later. The judgment result storage unit 12 stores the judgment result of the defect judgment unit 16, which will be described later, determining whether or not the defect candidate in the defect candidate region frontal image is a defect. The CPU 4 comprises the image extraction unit 13, the frontal viewpoint calculation unit 14, the image viewpoint conversion unit 15, and the defect judgment unit 16. The image extraction unit 13 acquires one captured image of the inspection area of the propeller 17 captured in one shot by the camera 2 stored in the image storage unit 6, and extracts the defect candidate region image from the captured image. Thus, since only one image is needed to capture the entire propeller 17 in a single shot, the imaging time can be shortened. Furthermore, the image extraction unit 13 corrects the brightness value of the defect candidate region image if it is below a predetermined value. This makes it possible to determine defects in the defect candidate region image that are difficult to see, even when the defect candidate region image is in a dark area, thereby improving the accuracy of defect determination.The frontal viewpoint calculation unit 14 calculates a frontal viewpoint facing the defect candidate region image using the coordinate information of the distance between the camera 2 and the inspection area of the propeller 17 stored in the coordinate information storage unit 7 and the propeller shape information stored in the object shape information storage unit 8. The image viewpoint conversion unit 15 converts the defect candidate region image into a frontal view image of the defect candidate region as seen from the frontal viewpoint. When the image viewpoint conversion unit 15 converts the defect candidate region image into a frontal view image of the defect candidate region as seen from the frontal viewpoint, it changes the resolution of the defect candidate region image. By changing the resolution of the defect candidate region image to a high resolution, the defect candidate shape can be measured with high accuracy, and by changing the resolution of the defect candidate region image to a low resolution, the defect candidate shape can be measured at high speed. The defect determination unit 16 determines whether or not the defect candidate in the frontal view image of the defect candidate region is a defect. Details of the processing of the frontal viewpoint calculation unit 14 and the defect determination unit 16 will be described later.
[0010] Figure 2 is the main flowchart showing the image inspection process in Embodiment 1.
[0011] In step S1, the image extraction unit 13 acquires a single image of the inspection area of the propeller 17 captured by camera 2 in one shot and stored in the image storage unit 6. In step S2, the image extraction unit 13 extracts a defect candidate area image from the acquired single image of the inspection area of the propeller 17, and corrects the brightness value of the defect candidate area image if it is less than or equal to a predetermined value. In step S3, the frontal viewpoint calculation unit 14 calculates a frontal viewpoint that is directly facing the defect candidate area image using coordinate information of the distance between camera 2 and the inspection area of the propeller 17 stored in coordinate information storage unit 7 and shape information of the propeller 17 stored in object shape information storage unit 8. In step S4, the image viewpoint conversion unit 15 changes the resolution of the defect candidate area image and converts it into a frontal view image of the defect candidate area as seen from the frontal viewpoint. In step S5, the defect determination unit 16 determines whether or not the defect candidate in the frontal view image of the defect candidate area is a defect.
[0012] Figure 3 is a time chart showing the calculation of the frontal viewpoint by the frontal viewpoint calculation unit in Embodiment 1.
[0013] In the first block Q1, the frontal viewpoint calculation unit 14 uses the set camera position coordinates ZK and propeller position coordinates Z, along with coordinate information of the distance between the camera 2 and the inspection area of the propeller 17, camera parameter information (focal length, etc.), and the position coordinates of the defect candidate area KR where the defect candidate K exists in the captured image, to geometrically identify the defect candidate area mesh face group M of the defect candidate area KR using 3D CAD data. In the second block Q2, the frontal viewpoint calculation unit 14 extracts the defect candidate area mesh face group M of the defect candidate area KR. The mesh data (STL format) of the 3D CAD data includes not only the 3D position information of each vertex of the triangle of each mesh, but also vector data representing the normal direction of each mesh. Furthermore, the size of each mesh in the mesh data (STL format) of the 3D CAD data is set to be less than or equal to the defect length threshold L0 (defect detection size). This makes it possible to determine defects in the defect candidate area image of the propeller 17 with high accuracy. In the third block Q3, the frontal viewpoint calculation unit 14 obtains the mesh surface normal H of each mesh. In the fourth block Q4, the frontal viewpoint calculation unit 14 approximates the origin G of the defect candidate surface normal and the defect candidate surface normal H0 (the frontal viewpoint S1 of the defect candidate region, described later) from the position on the vector obtained by combining the intersection points of the mesh surface normals H of each mesh. As a result, the defect candidate K can be viewed from the frontal viewpoint, making it possible to accurately measure the size of the defect candidate K, preventing defects from being overlooked, and improving the accuracy of defect detection.
[0014] Figure 4 is a time chart showing the process flow for defect determination in Embodiment 1.
[0015] In the first block P1, the image storage unit 6 acquires and stores the captured image. In the second block P2, the image extraction unit 13 extracts a defect candidate region image from the captured image stored in the image storage unit 6. In the third block P3, the frontal viewpoint calculation unit 14 uses coordinate information of the distance between the camera 2 and the inspection area of the propeller 17, camera parameter information (focal length, etc.), and the position coordinates of the defect candidate region KR in the captured image where the defect candidate K exists to roughly calculate the frontal viewpoint S1 of the defect candidate region relative to the camera 2's imaging viewpoint S. In the fourth block P4, the image viewpoint conversion unit 15 converts the defect candidate image into a frontal view image of the defect candidate as seen from the frontal viewpoint S1. This allows the defect candidate K to be converted to its actual size, improving the defect determination accuracy. In the fifth block P5, the defect determination unit 16 determines whether the maximum length L of the defect candidate K in the frontal view image of the defect candidate is greater than or equal to the length threshold L0 or less, and determines that it is a defect if the maximum length L of the defect candidate K is greater than or equal to the length threshold L0. This improves the defect determination accuracy.
[0016] Next, the effects of Embodiment 1 will be explained.
[0017] (1) The image inspection device 1 includes an image storage unit 6 that acquires and stores an image of the inspection area of a propeller 17 having a three-dimensional curved surface captured by a camera 2; a coordinate information storage unit 7 that stores coordinate information of the distance between the camera 2 and the inspection area of the propeller 17; an object shape storage unit 8 that stores the shape information of the camera 2 and the propeller 17 having a three-dimensional curved surface; an image extraction unit 13 that extracts a defect candidate area image from the captured image; a frontal viewpoint calculation unit 14 that calculates a frontal viewpoint S1 (defect candidate surface normal H0) that is directly facing the defect candidate area image using the coordinate information of the distance between the camera 2 and the inspection area of the propeller 17 and the shape information of the propeller 17; an image viewpoint conversion unit 15 that converts the defect candidate area image into a frontal view of the defect candidate area as seen from the frontal viewpoint S1 (defect candidate surface normal H0); and a defect determination unit 16 that determines whether or not a defect candidate K in the frontal view of the defect candidate area is a defect. Therefore, accurate size measurement becomes possible even for defects on the curved surface of the propeller 17, which has a three-dimensional curved surface, thus preventing defects from being overlooked and improving the accuracy of defect detection.
[0018] (2) The shape information of the propeller 17 of the object under inspection shape memory unit 8 is in the form of 3D CAD data (STL format) with a mesh surface. Therefore, the 3D CAD data (STL format) includes not only the 3D position information of each vertex of the mesh triangle, but also vector data representing the normal direction of each mesh, making it easy to calculate the direct viewpoint S1 (normal H0 of the defect candidate surface), and eliminating the need to measure with a 3D measuring machine, thereby reducing equipment costs.
[0019] (3) The direct viewpoint S1 (defect candidate surface normal H0) calculated by the direct viewpoint calculation unit 14 is the direct viewpoint S1 (defect candidate surface normal H0) at a position directly facing the mesh surface of the 3D CAD data (STL format). Therefore, it is possible to make it easier to calculate the direct viewpoint S1 (defect candidate surface normal H0).
[0020] (4) The direct viewpoint S1 (defect candidate surface normal H0) calculated by the direct viewpoint calculation unit 14 is located on a vector obtained by combining normals perpendicular to multiple mesh surfaces of the 3D CAD data (STL format). Therefore, it is possible to make it easier to calculate the direct viewpoint S1 (defect candidate surface normal H0) and to calculate it with higher accuracy than if it were calculated on a single mesh surface.
[0021] (5) The mesh size of each mesh in the mesh data (STL format) of the 3D CAD data of the shape information of the propeller 17 of the object under inspection shape memory unit 8 is set to be less than or equal to the defect length threshold L0 (defect detection size). Therefore, it is possible to determine defects in candidate defects within the defect candidate region image of the propeller 17 with high accuracy.
[0022] (6) The image extraction unit 13 corrects the brightness value of the defect candidate region image if it is below a predetermined value. Therefore, when the defect candidate region image is in a dark position, it is possible to determine defects in the defect candidate region image that are difficult to see, thereby improving the accuracy of defect determination.
[0023] (7) When the image viewpoint conversion unit 15 converts the defect candidate region image into a defect candidate region image viewed from a frontal viewpoint, it changes the resolution of the defect candidate region image. Therefore, by changing the resolution of the defect candidate region image to a high resolution, the defect candidate shape can be measured with high accuracy, and by changing the resolution of the defect candidate region image to a low resolution, the defect candidate shape can be measured at high speed.
[0024] (8) The defect determination unit 16 determines whether the maximum length L of the defect candidate K in the defect candidate image is greater than or equal to the length threshold L0 or less, and determines that it is a defect if the maximum length L of the defect candidate K is greater than or equal to the length threshold L0. Therefore, it can handle inspections that determine whether something is a defect based on the maximum length of the defect candidate.
[0025] [Embodiment 2] Figure 5 is a time chart showing the calculation of the frontal viewpoint by the frontal viewpoint calculation unit in Embodiment 2.
[0026] In Embodiment 1, the direct viewpoint calculation unit 14 approximates the origin of the defect candidate surface normal G and the defect candidate surface normal H0 (direct viewpoint S1) from the position on a vector obtained by combining the intersection points of the mesh surface normals H of each mesh. In Embodiment 2, however, as shown in the third block Q3a, the direct viewpoint calculation unit 14 extracts boundary vectors J at the boundaries of each mesh surface where the defect candidate K overlaps, and as shown in the fourth block Q4a, calculates perpendiculars S from pairs of adjacent boundary vectors J by vector cross product calculation, and approximates the defect candidate surface normal H0 and the origin of the defect candidate surface normal G from the average value of multiple perpendiculars S.
[0027] Since the other components are the same as those in Embodiment 1, the same reference numerals are used for the same components, and their descriptions are omitted. Therefore, Embodiment 2 provides the same effects and advantages as Embodiment 1.
[0028] [Embodiment 3] Figure 6 is an overall schematic diagram of the image inspection apparatus 1 of Embodiment 3.
[0029] In Embodiment 1, a camera 2 with a photographic lens that exhibits low lens distortion was used, but in Embodiment 3, a camera 2a with high image distortion, such as a fisheye or 360-degree camera, is used.
[0030] Since the other components are the same as those in Embodiment 1, the same reference numerals are used for the same components, and their descriptions are omitted. Thus, in Embodiment 3, in addition to the effects of Embodiment 1, the effect of being able to widen the imaging range of the captured image is achieved.
[0031] [Embodiment 4] Figure 7 is an overall schematic diagram of the image inspection apparatus 1 of Embodiment 4.
[0032] In Embodiment 1, a camera 2 with a photographic lens that has low lens distortion was used, and the shape information of the object under inspection, including the dimensional information of the propeller 17 (design dimensions, shape data, etc.) that had been prepared in advance, was stored in the object under inspection shape information storage unit 8. However, in Embodiment 4, a camera 2b capable of three-dimensional measurement is used, and the shape information of the propeller 17 is obtained from the image captured by this camera 2b and stored in the object under inspection shape information storage unit 8a.
[0033] Since the other components are the same as those in Embodiment 1, the same reference numerals are used for the same components, and their descriptions are omitted. Thus, in Embodiment 3, in addition to the effects of Embodiment 1, the effect of enabling the determination of defects in candidate defects is achieved even if the shape of the object to be inspected, such as the propeller 17, and the coordinate information of the distance between the camera 2b and the inspection area are not known.
[0034] [Embodiment 5] Figure 8 is an overall schematic diagram of the image inspection apparatus 1 of Embodiment 5.
[0035] In Embodiment 1, the object shape information of the object to be inspected, including dimensional information (design dimensions, shape data, etc.) of the propeller 17 prepared in advance, was stored in the object shape information storage unit 8. However, in Embodiment 5, the object shape information storage unit 8b acquires and stores the object shape information estimated from the captured image using a 3D shape estimation process such as AI.
[0036] Since the other components are the same as those in Embodiment 1, the same reference numerals are used for the same components, and their descriptions are omitted. Thus, in Embodiment 5, in addition to the effects of Embodiment 1, shape information can be obtained from the captured image of the inspection area of the object under inspection by 3D shape estimation processing such as AI, and it is possible to determine defects in candidate defects even without shape information from CAD or 3D measuring devices.
[0037] [Embodiment 6] Figure 9 is a time chart showing the process flow for defect determination in Embodiment 6.
[0038] In Embodiment 1, the defect determination unit 16 determined whether the maximum length L of the defect candidate K in the defect candidate image was greater than or equal to the length threshold L0, and determined that it was a defect if the maximum length L of the defect candidate K was greater than or equal to the length threshold L0. In Embodiment 6, however, the defect determination unit 16 determined whether the area A of the defect candidate K in the defect candidate image was greater than or equal to the area threshold A0, and determined that it was a defect if the area A of the defect candidate K was greater than or equal to the area threshold A0.
[0039] Since the other components are the same as those in Embodiment 1, the same reference numerals are used for the same components, and their descriptions are omitted. Thus, Embodiment 6 provides the effect of being able to perform inspections that determine whether a defect is present based on the area of the defect candidate, instead of the effect (8) of Embodiment 1.
[0040] [Other Embodiments] Although embodiments for carrying out the present invention have been described above, the specific configuration of the present invention is not limited to the configurations of the embodiments, and design changes and the like that do not depart from the gist of the invention are also included in the present invention. The object to be inspected is not limited to a propeller, and may be anything having a three-dimensional curved surface.
[0041] It should be noted that the present invention is not limited to the embodiments described above, and various modifications are included. For example, the embodiments described above are described in detail to make the present invention easier to understand, and are not necessarily limited to those having all the configurations described. Furthermore, it is possible to replace parts of the configuration of one embodiment with the configuration of another embodiment, and it is also possible to add configurations from other embodiments to the configuration of one embodiment. In addition, it is possible to add, delete, or replace parts of the configuration of each embodiment with other configurations.
[0042] This application claims priority under Japanese Patent Application No. 2025-014496, filed on 31 January 2025. The entire disclosure of Japanese Patent Application No. 2025-014496, filed on 31 January 2025, including the specification, claims, drawings, and abstract, is incorporated into this application by reference.
[0043] 1 Image inspection device, 2 Camera with a photographic lens with low lens distortion (imaging device), 2a Camera with high image distortion such as a fisheye or 360-degree camera (imaging device), 2b Camera capable of 3D measurement (imaging device), 3 Computer, 4 Memory, 5 CPU, 6 Image storage unit, 7 Coordinate information storage unit, 8, 8a, 8b Inspection object shape information storage unit, 13 Image extraction unit, 14 Frontal viewpoint calculation unit, 15 Image viewpoint conversion unit, 16 Defect determination unit, 17 Propeller (inspection object), A Defect candidate area, A0 Area threshold, L Defect candidate maximum length, L0 Length threshold, K Defect candidate, KR Defect candidate region, M Mesh surface group, S Imaging viewpoint, S1 Frontal viewpoint
Claims
An image inspection device for inspecting defects in an inspection area of an object having a three-dimensional curved surface, An image storage unit that acquires and stores an image captured by an imaging device of the inspection area of the object to be inspected, A coordinate information storage unit that stores coordinate information of the distance between the imaging device and the inspection area, A storage unit for object shape information that stores the shape information of the object to be inspected as object shape information, An image extraction unit that extracts defect candidate region images from the aforementioned captured image, A frontal viewpoint calculation unit calculates a frontal viewpoint that is directly facing the defect candidate region image using the coordinate information and the shape information of the object to be inspected, An image viewpoint conversion unit converts the defect candidate region image into a defect candidate region image viewed from the correct viewpoint, A defect determination unit that determines whether or not a defect candidate in the defect candidate region image is a defect, An image inspection device having the following features. An image inspection apparatus according to claim 1, The shape information of the object to be inspected is 3D CAD data. Image inspection device. An image inspection apparatus according to claim 2, The aforementioned direct viewpoint is a viewpoint positioned directly opposite the mesh plane of the 3D CAD data of the shape information of the object under inspection. Image inspection device. An image inspection apparatus according to claim 3, The aforementioned direct viewpoint lies on a vector obtained by combining normals perpendicular to multiple mesh planes of the 3D CAD data of the object shape information under inspection. Image inspection device. An image inspection apparatus according to claim 3 or 4, The mesh size of the 3D CAD data of the object shape information is set to be less than or equal to the defect detection size. Image inspection device. An image inspection apparatus according to claim 1, The image extraction unit corrects the brightness value of the defect candidate region image if the brightness value of the defect candidate region image is below a predetermined value. Image inspection device. An image inspection apparatus according to claim 1, The aforementioned imaging device is a camera with significant image distortion, such as a fisheye or 360-degree camera. The aforementioned frontal viewpoint calculation unit calculates the frontal viewpoint using the image captured by the camera. Image inspection device. An image inspection apparatus according to claim 1, The aforementioned shape information of the object to be inspected is shape information of the object to be inspected obtained from an image captured by a camera capable of three-dimensional measurement. The aforementioned direct viewing viewpoint calculation unit calculates the direct viewing viewpoint using the shape information of the object to be inspected. Image inspection device. An image inspection apparatus according to claim 1, The aforementioned object shape information is the object shape information estimated from the captured image by a three-dimensional shape estimation process. The aforementioned direct viewing viewpoint calculation unit calculates the direct viewing viewpoint using the shape information of the object to be inspected. Image inspection device. An image inspection apparatus according to claim 1, The image viewpoint conversion unit changes the resolution of the defect candidate region image when converting the defect candidate region image into a defect candidate region image. Image inspection device. An image inspection apparatus according to claim 1, The defect determination unit determines a defect based on the maximum length of the defect candidate within the defect candidate region image. Image inspection device. An image inspection apparatus according to claim 1, The defect determination unit determines a defect based on the area of the defect candidate within the defect candidate region image. Image inspection device.