Enhanced industrial visual detection method based on artificial intelligence

Abstract

The invention discloses an enhanced industrial visual inspection method based on artificial intelligence, and belongs to the technical field of machine visual analysis. According to the appearance inspection method, a plurality of industrial cameras are arranged along conveying equipment to collect appearance images of a rigid object at multiple angles; and acquiring and analyzing the image of the same rigid object with the similar angle in the historical record according to the angle corresponding to the acquired image, thereby judging whether the image acquired in real time is abnormal or not, and compared with a traditional visual analysis scheme, the method has lower requirements on the placement position and the placement angle of the rigid object, and the accuracy of the visual analysis is improved. The rigid objects to be detected do not need to be subjected to fixed transmission detection one by one and then subjected to positioning removal one by one, so that the visual analysis cost can be remarkably reduced, and the analysis efficiency is improved.

Description

Technical Field

[0001] The present invention belongs to the field of machine vision technology, and specifically relates to an enhanced industrial vision inspection method based on artificial intelligence. Background Art

[0002] With the advancement and development of science and technology, intelligent manufacturing has become an important development direction of modern manufacturing. Among them, using machine vision technology to perform appearance inspection of products to identify defects on the product surface is a common intelligent manufacturing method. Compared with manual visual inspection, it is more efficient, has lower labor costs and higher accuracy, and has therefore been widely used.

[0003] When using machine vision technology to perform appearance inspection on products, for objects with complex surface shapes, the object is generally positioned, and then an image of the corresponding surface of the object is captured by a fixed camera, and then the image analysis is used to determine whether there are any defects in the appearance of the object. However, this requires that the object be positioned before appearance inspection. Many objects are not suitable for one-to-one fixed transmission due to modification costs and their own production characteristics. Therefore, this results in different angles of the object during transmission, and the images captured by the same camera also correspond to different positions of the object, which makes appearance inspection through machine vision more difficult. In order to solve the above problems, a method for accurate appearance inspection is provided without fixing the position of the object to be inspected. The present invention provides the following technical solutions. Summary of the Invention

[0004] The purpose of the present invention is to provide an enhanced industrial vision inspection method based on artificial intelligence to solve the problem that in the prior art, when using machine vision technology to inspect the appearance of products, for objects with complex surface shapes, the method generally positions the object, then uses a fixed camera to capture an image of the corresponding surface of the object, and uses image analysis to determine whether there are any defects in the appearance of the object, which requires high positioning of rigid objects.

[0005] The purpose of the present invention can be achieved through the following technical solutions:

[0006] The enhanced industrial visual inspection method based on artificial intelligence includes the following steps:

[0007] In the first step, when a rigid object is transported on a conveyor, each industrial camera collects images of the rigid object's surface in sequence.

[0008] In the second step, an image of a rigid object surface captured in real time by an industrial camera is marked as the measured image;

[0009] Searching the database for surface images of the same rigid object, obtaining a surface image whose similarity to the image being measured is greater than a preset proportional coefficient α and marking it as a comparison image;

[0010] The third step is to obtain the point cloud of each comparison image, and then segment the plane based on the randomly sampled consistent point cloud;

[0011] The fourth step is to divide the images of the same plane in each comparison image into a plane image control group;

[0012] Obtaining grayscale images of each image in a plane image control group, and dividing these grayscale images into a number of grid areas according to the same rule;

[0013] Get the grayscale value average u corresponding to each grayscale image in the same grid area, then calculate the average value up of these grayscale value averages u, and mark up as the grayscale parameter of the corresponding grid area;

[0014] The fifth step is to obtain the grayscale average value us of the grayscale image in the corresponding square area of the measured image. When the difference between us and up is greater than the threshold, it is considered that there is an abnormality in the corresponding square area of the measured image.

[0015] Furthermore, the industrial cameras are arranged along the laying direction of the conveying equipment, and each industrial camera has a different acquisition angle corresponding to the rigid object.

[0016] Furthermore, in the third step, the edge lines of each plane in each comparison image are obtained according to the segmented planes, and the ratio of each comparison image and the measured image is adjusted according to the positioning of the edge lines and the intersection points between the edge lines.

[0017] Furthermore, in the fourth step, when |u-up| / up>θ exists among the obtained u values, the corresponding u value is deleted;

[0018] Where θ is the preset proportional coefficient.

[0019] Furthermore, in the fifth step, when |us-up| / up>θ1 holds, it is considered that the image under test has an anomaly in the corresponding grid area;

[0020] Where θ1 is the preset proportional coefficient.

[0021] Furthermore, after the fifth step, the distribution of abnormal grid areas on each plane in the measured image is obtained. When the number of abnormal grid areas within a preset range is greater than a preset threshold, the appearance of the plane is considered abnormal, and the corresponding rigid object is marked as an abnormal object.

[0022] Beneficial effects of the present invention:

[0023] The appearance test method described in the present invention collects appearance images of rigid objects at multiple angles by setting up multiple industrial cameras along the conveying equipment, and obtains and analyzes images of the same rigid object at similar angles in the historical records based on the angles corresponding to the collected images, so as to judge whether there are abnormalities in the real-time collected images. Compared with traditional visual analysis solutions, the requirements for the placement position and placement angle of rigid objects are lower, so it can significantly reduce the cost of visual analysis and improve analysis efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The present invention will be further described below with reference to the accompanying drawings.

[0025] Figure 1 1 is a flow chart of an enhanced industrial visual inspection method based on artificial intelligence according to the present invention;

[0026] Figure 2 It is a framework diagram of an industrial vision inspection system based on an enhanced industrial vision inspection method based on artificial intelligence. DETAILED DESCRIPTION

[0027] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0028] Example 1

[0029] Enhanced industrial visual inspection methods based on artificial intelligence, such as Figure 1 As shown, the following steps are included:

[0030] In the first step, the rigid object is transferred to a conveyor device via a transfer device. Several industrial cameras are set up along the laying direction of the conveyor device. Each industrial camera has a different acquisition angle corresponding to the rigid object, that is, each industrial camera captures a different image of the rigid object;

[0031] The number of industrial camera settings is related to the surface shape of the rigid object. The more planes the surface corresponds to, the more industrial camera settings there are.

[0032] When a rigid object is transported on the conveying equipment, each industrial camera collects images of the rigid object's surface in sequence;

[0033] The second step is to mark the surface image of a rigid object captured by an industrial camera. The mark can distinguish the industrial camera and the functional object corresponding to the corresponding surface image.

[0034] For easier understanding, here are some examples:

[0035] The surface image of a rigid object captured by an industrial camera is labeled Pij, where 1≤i≤n, n is the number of the corresponding industrial camera, and j is the number of the corresponding rigid object;

[0036] The third step is to use a surface image of a rigid object captured in real time by an industrial camera as the research object for subsequent steps. For the convenience of description, the surface image captured in real time is called the measured image.

[0037] Searching the database for surface images of the same rigid object, obtaining surface images whose similarity to the measured image is greater than a preset proportional coefficient α. For the convenience of description, these surface images are referred to as comparison images.

[0038] The fourth step is to obtain the point cloud of each comparison image, and then segment the plane based on the randomly sampled consistent point cloud to obtain the edge lines of each plane in each comparison image. The ratio of each comparison image to the measured image is adjusted based on the edge lines and the intersection points between the edge lines.

[0039] The fifth step is to summarize the planes segmented in each comparison image, that is, to divide the images of the same plane in each comparison image into a plane image control group;

[0040] Obtaining grayscale images of each image in a plane image control group, and dividing these grayscale images into a number of grid areas according to the same rule;

[0041] Get the grayscale value average u corresponding to each grayscale image in the same grid area, then calculate the average value up of these grayscale value averages u, and mark up as the grayscale parameter of the corresponding grid area;

[0042] It should be noted that in this step, when |u-up| / up>θ exists among the obtained u values, the corresponding u value will be deleted to ensure the accuracy of the results and reduce the influence of accidental factors;

[0043] Where θ is the preset proportional coefficient;

[0044] For easier understanding, here are some examples:

[0045] The number of planes of the object under test in a comparison image that can be used for subsequent analysis is 3. These three planes are marked as q1, q2, and q3 in sequence. Then, the images of plane q1 in all comparison images are divided into a plane image control group, and the subsequent grayscale parameter calculation is performed;

[0046] Step 6: Obtain the grayscale average value us of the grayscale image in the corresponding grid area of the measured image, and compare us with up. If the difference between the two is large, it is considered that there is an abnormality in the corresponding grid area of the measured image;

[0047] Specifically, when |us-up| / up>θ1 holds, it is considered that the image under test has an anomaly in the corresponding grid area;

[0048] Where θ1 is the preset proportional coefficient;

[0049] Step 7: Obtain the distribution of abnormal grid areas on each plane of the measured image. When the number of abnormal grid areas within a certain range exceeds a preset threshold, the plane is considered to have an abnormal appearance, and the corresponding rigid object is marked as an abnormal object. Subsequent operations such as removal and visual inspection are performed;

[0050] Specifically, a rectangular range is arbitrarily defined. When the number of grid areas within the range exceeds a preset threshold, the corresponding rigid object is marked as an abnormal object, where the preset threshold is related to parameters such as the size of the grid area.

[0051] This appearance test method collects appearance images of rigid objects at multiple angles by setting up multiple industrial cameras along the conveying equipment. Based on the angles corresponding to the collected images, images of the same rigid object at similar angles are acquired and analyzed in historical records, thereby judging whether there are any abnormalities in the real-time collected images. Compared with traditional visual analysis solutions, this method has lower requirements on the placement position and placement angle of rigid objects, thus significantly reducing the cost of visual analysis and improving analysis efficiency.

[0052] Example 2

[0053] This embodiment discloses an industrial visual inspection system for executing the artificial intelligence-based enhanced industrial visual inspection method in the above embodiment 1, such as Figure 2 As shown, the system includes:

[0054] A conveying device for conveying rigid objects, wherein the conveying device is preferably arranged in a straight line in the area of the appearance test (i.e., the area where the industrial camera is set);

[0055] Industrial cameras are laid along the conveying equipment, and each industrial camera has a different shooting angle. The industrial cameras collect images of rigid objects at multiple angles and transmit the collected images to the retrieval unit;

[0056] a retrieval unit, configured to retrieve images similar to an image of a rigid object captured by an industrial camera from a database, and transmit the retrieved images to a comparison unit;

[0057] A database for storing images of rigid objects captured by industrial cameras that do not have abnormal appearances;

[0058] In order to avoid wasting computing power, the amount of image information stored in the database can be limited, which can reduce storage and retrieval costs without affecting the adequacy of samples.

[0059] A comparison unit is used to compare the rigid object image acquired in real time by the industrial camera with the plurality of images retrieved by the retrieval unit, thereby identifying whether there is any abnormality in the rigid object image acquired in real time;

[0060] In addition, when an image is determined to be abnormal by the comparison unit, the image will not be recorded in the database.

[0061] The above contents are merely examples and explanations of the present invention. Those skilled in the art may make various modifications or additions to the described specific embodiments or replace them in similar ways. As long as they do not deviate from the invention or exceed the scope defined by the claims, they should all fall within the scope of protection of the present invention.

Claims

1. An enhanced industrial visual inspection method based on artificial intelligence, characterized in that: The steps include: In the first step, when a rigid object is transported on a conveyor, each industrial camera collects images of the rigid object's surface in sequence. In the second step, an image of a rigid object surface captured in real time by an industrial camera is marked as the measured image; Searching the database for surface images of the same rigid object, obtaining a surface image whose similarity to the image being measured is greater than a preset proportional coefficient α and marking it as a comparison image; The third step is to obtain the point cloud of each comparison image, and then segment the plane based on the randomly sampled consistent point cloud; The fourth step is to divide the images of the same plane in each comparison image into a plane image control group; Obtaining grayscale images of each image in a plane image control group, and dividing these grayscale images into a number of grid areas according to the same rule; Get the grayscale value average u corresponding to each grayscale image in the same grid area, then calculate the average value up of these grayscale value averages u, and mark up as the grayscale parameter of the corresponding grid area; The fifth step is to obtain the grayscale average value us of the grayscale image in the corresponding square area of the measured image. When the difference between us and up is greater than the threshold, it is considered that there is an abnormality in the corresponding square area of the measured image.

2. The artificial intelligence-based enhanced industrial visual inspection method according to claim 1, characterized in that: The industrial cameras are set along the laying direction of the conveying equipment, and each industrial camera has a different acquisition angle corresponding to the rigid object.

3. The artificial intelligence-based enhanced industrial visual inspection method according to claim 1, characterized in that: In the third step, the edge lines of each plane in each comparison image are obtained according to the segmented planes, and the ratio of each comparison image and the measured image is adjusted according to the positioning of the edge lines and the intersection points between the edge lines.

4. The artificial intelligence-based enhanced industrial visual inspection method according to claim 1, characterized in that: In the fourth step, when |u-up| / up>θ exists among the obtained u values, the corresponding u value is deleted; Where θ is the preset proportional coefficient.

5. The artificial intelligence-based enhanced industrial visual inspection method according to claim 4, characterized in that: In the fifth step, when |us-up| / up>θ1, it is considered that the image under test has an anomaly in the corresponding grid area; Where θ1 is the preset proportional coefficient.

6. The artificial intelligence-based enhanced industrial visual inspection method according to claim 1, characterized in that: After the fifth step, the distribution of abnormal grid areas on each plane in the measured image is also obtained. When the number of abnormal grid areas within a preset range is greater than a preset threshold, the appearance of the plane is considered abnormal, and the corresponding rigid object is marked as an abnormal object.

Images