Cambered surface visual inspection industrial camera and cambered surface visual inspection method thereof

Abstract

The invention discloses a cambered surface visual inspection industrial camera and a cambered surface visual inspection method thereof, and belongs to the technical field of industrial cameras. The cambered surface visual detection method comprises the following steps: randomly selecting a p point in an image, the brightness value of the pixel of the p point being Ip; judging whether the p point is the central point of the angular point or not: setting a threshold value t, comparing the gray value difference between the p point and the pixel points around the p point according to the threshold value t, forming a circle by taking the p point as the circle center and r as the radius, covering M pixels around the p point, setting r to be 3, then M to be 16, i.e., 16 pixel points exist on the circumference, and comparing the gray values of the 16 pixel points on the circumference, so as to determine whether the p point is the central point of the angular point or not. Pixels with the gray values larger than Ip + t and smaller than Ip-t are obtained, and if n continuous pixels meet the condition that all the pixels are larger than Ip + t or all the pixels are smaller than Ip-t, the p point is regarded as an angular point. By means of the light source array, the lens, the photosensitive element and the deep learning algorithm which are specially designed, high-precision detection of the cambered surface object in the closed narrow space can be achieved, and the existing technical problems are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial cameras, and particularly to a spherical vision detection industrial camera and a spherical vision detection method thereof. Background Art

[0002] Since the concept of machine vision was proposed in the 1950s and after nearly seventy years of development, machine vision has become an important branch of artificial intelligence. Machine vision systems relying on key technologies such as circuit technology, semiconductor technology, and operating systems are widely used in fields such as factory automation, medical and life sciences, retail, transportation, security monitoring, and logistics. The wide application of machine vision systems has promoted the development process of industrial cameras.

[0003] However, at present, machine vision technologies all use natural light or artificial supplementary light as the light source. The instability of the light source causes machine vision technology not to be applicable in some specific enclosed and narrow spaces, seriously hindering the development of machine vision technology.

[0004] In the prior art, industrial cameras are usually used in cooperation with bar-shaped light sources or ring-shaped light sources. The object to be measured is usually located in an open space, and the arrangement of the camera and the light source is relatively simple. Although the independent light source is equipped with a dedicated controller, the system composed of it and a general industrial camera has a large volume and is mainly used for visual analysis of flat or regular-shaped objects, but it cannot meet the narrow and enclosed space. For irregular spherical surfaces such as the inner cavity of a cylinder, the engine housing, and the inner wall of a valve body, the diversity of their surface shapes poses great challenges to detection. At the same time, due to the limitation of the optical environment in the enclosed space, the light source configuration also becomes a major problem. The above reasons result in the traditional detection method being unable to meet the application requirements and being difficult to detect the surface and dimensions that are difficult for the human eye to reach, resulting in a decrease in detection accuracy. Summary of the Invention

[0005] The purpose of the present invention is to provide a spherical vision detection industrial camera and a spherical vision detection method thereof. Through a specially designed light source array, lens, photosensitive element, and deep learning algorithm, the present invention can achieve high-precision detection of spherical objects in narrow and enclosed spaces and solve the problems of the prior art.

[0006] The present invention is realized through the following technical solutions:

[0007] The present invention provides a spherical vision detection method, including:

[0008] Arbitrarily select a p point in the image, and the brightness value of the pixel at the p point is Ip.

[0009] Judging whether the p point is the center point of a corner point includes:

[0010] Set a threshold t, which compares the grayscale value difference between point p and its surrounding pixels. A circle is formed with point p as the center and r as the radius, covering M pixels around point p. If r=3, then M=16, that is, there are 16 pixels on the circumference.

[0011] The grayscale values ​​of the 16 pixels on the circumference are compared to obtain pixels whose grayscale values ​​exceed Ip+t and are lower than Ip-t. If there are n consecutive pixels that satisfy the condition that they are all greater than Ip+t or less than Ip-t, then point p is considered to be a corner point.

[0012] Furthermore, the fast detection of corner points in the image includes:

[0013] Detect the grayscale values ​​of the four pixels numbered 1, 9, 5, and 13 on the circumference. If 3 or more pixels among these four points meet the condition that they are all greater than Ip+t or less than Ip-t, then the point p is a corner point, otherwise it is directly excluded.

[0014] Furthermore, to avoid the concentration of corner points, non-maximum suppression is performed including:

[0015] In a certain area, retain the corner points with the maximum response value.

[0016] The corner point of the response maximum value is obtained by calculating the maximum value of the minimum value of all consecutive n pixels in the absolute difference between the corner point and the 16 pixels on its circumference.

[0017] If point p satisfies the grayscale values ​​of n consecutive pixels and is still retained after non-maximum suppression, then the point is determined to be a corner point.

[0018] Furthermore, it also includes a DNN module, which imports and runs models trained under different deep learning frameworks, and uses the FAST feature extraction algorithm by combining the OpenCV with the deep learning framework.

[0019] Comparing the grayscale values ​​of a pixel with its surrounding neighborhood pixels is achieved through the FAST feature extraction algorithm.

[0020] A curved surface visual inspection industrial camera comprises an industrial camera body (2), the industrial camera body (2) is provided with a camera lens (3), a ring-shaped curved concave light source (1) is provided on one side of the industrial camera body (2), the ring-shaped curved concave light source (1) is configured to be detachable, 16 LED lamp cores are evenly arranged on the concave surface of the ring-shaped curved concave light source (1), and a single-chip microcomputer is provided inside the industrial camera body (2), and the single-chip microcomputer independently controls each LED lamp core through pulse width.

[0021] Furthermore, the single chip microcomputer adjusts the duty cycle of the output signal by changing the pulse width of the control signal.

[0022] Furthermore, the control signal consists of a periodic square wave with a fixed frequency and a rectangular pulse with a variable duty cycle; the high level of the square wave represents a signal "1", and the low level represents a signal "0"; the width of the rectangular pulse represents the duration of the high level, and the duty cycle is determined by the proportion of the duration of the high level to the entire period; the brightness of the LED is proportional to the intensity of the current passing through it; through the PWM technology, the current intensity of the LED is precisely controlled, thereby achieving precise control of the brightness.

[0023] The present invention has the following beneficial effects:

[0024] 1. Through specially designed light source arrays, lenses, photosensitive elements and deep learning algorithms, the present invention can achieve high-precision detection of curved objects in closed and narrow spaces, solving the problems of the prior art.

[0025] 2. The system integration solution provided by the present invention can realize real-time and automatic visual analysis and improve detection efficiency due to the self-adaptation and autonomous control of the light source.

[0026] 3. The industrial camera provided by the present invention adopts an embedded control hardware architecture, and the software supports multiple signal output modes, which can achieve direct replacement with the existing system.

[0027] 4. The visual inspection system of the present invention has a self-diagnosis function, which can monitor the system status and diagnose faults in a timely manner.

[0028] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 This is a schematic diagram of the industrial camera structure;

[0030] Figure 2 This is a schematic diagram of the side view structure of an industrial camera;

[0031] Figure 3 The figure is the flow chart of the integrated system.

[0032] In the figure: 1. Annular arc concave light source; 2. Industrial camera body; 3. Camera lens. DETAILED DESCRIPTION

[0033] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. 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 creative work are within the scope of protection of the present invention.

[0034] See also Figure 1-3 The present invention provides a technical solution: a curved surface visual detection method, comprising:

[0035] Select any point p in the image, the brightness value of the pixel at point p is Ip.

[0036] Judging whether point p is the center point of a corner point includes:

[0037] Set a threshold t, which compares the grayscale value difference between point p and its surrounding pixels. A circle is formed with point p as the center and r as the radius, covering M pixels around point p. If r=3, then M=16, that is, there are 16 pixels on the circumference.

[0038] First, the pixels to be detected (candidate points) are selected from the input image. Usually these points are each pixel of the image. For each candidate point, a circular neighborhood around the point is considered, with a radius of 3 pixels and containing 16 pixels.

[0039] The grayscale values ​​of the 16 pixels on the circumference are compared to obtain pixels with grayscale values ​​greater than Ip+t and less than Ip-t. If there are n consecutive pixels (n is usually 12, i.e. FAST-12; other commonly used n values ​​are 9 and 11, known as FAST-9 and FAST-11, respectively) that meet the conditions of being greater than Ip+t or less than Ip-t, then point p is considered to be a corner point.

[0040] The role of determining corner points is very important in computer vision, especially in the fields of image processing, image analysis, target recognition, and video surveillance. Corner points are points in an image where the local structure changes significantly, usually located at the edges and corners of an object. Because they exhibit unique local structures in images, they are often used as feature points for tasks such as image matching, tracking, and 3D reconstruction.

[0041] Corner points play a vital role in computer vision, especially in applications that require accurate positioning and matching of image features. Since corner points have high recognition and stability in images, they are often used as the basis for subsequent processing (such as tracking, reconstruction, recognition, etc.). Through the effective detection and utilization of corner points, many visual tasks can be successfully completed.

[0042] In the detection of surface defects of workpieces, the industrial camera transmits the captured images to the host computer, and the algorithm in the host computer identifies the corner points, that is, locates the defects of the photographed workpiece, such as ruptures, dents, bubbles and cracks, etc.

[0043] Fast detection of corner points in an image includes:

[0044] Detect the grayscale values ​​of the four pixels numbered 1, 9, 5, and 13 on the circumference. If 3 or more pixels among these four points meet the condition that they are all greater than Ip+t or less than Ip-t, then the point p is a corner point, otherwise it is directly excluded.

[0045] Avoiding corner point concentration and performing non-maximum suppression include:

[0046] In a certain area (such as a window of odd size such as 3×3, 5×5 or 7×7), the corner points with the maximum response value are retained.

[0047] The corner point of the response maximum value is obtained by calculating the maximum value of the minimum value of all consecutive n pixels in the absolute difference between the corner point and the 16 pixels on its circumference.

[0048] If point p satisfies the grayscale values ​​of n consecutive pixels and is still retained after non-maximum suppression, then the point is determined to be a corner point.

[0049] It also includes a DNN module, which imports and runs models trained under different deep learning frameworks, and uses the FAST feature extraction algorithm by combining the OpenCV with the deep learning framework.

[0050] Comparing the grayscale values ​​of a pixel with its surrounding neighborhood pixels is achieved through the FAST feature extraction algorithm.

[0051] The camera is used to take pictures, and then the pictures are uploaded to the image processing module inside the camera. The image processing module inside the camera uses the FAST feature extraction algorithm to process the pictures, thereby determining the positions of the corner points and completing the image processing to ensure that a clear image can be obtained.

[0052] A curved surface visual inspection industrial camera comprises an industrial camera body (2), the industrial camera body (2) is provided with a camera lens (3), a ring-shaped curved concave light source (1) is provided on one side of the industrial camera body (2), the ring-shaped curved concave light source (1) is configured to be detachable, 16 LED lamp cores are evenly arranged on the concave surface of the ring-shaped curved concave light source (1), and a single-chip microcomputer is provided inside the industrial camera body (2), and the single-chip microcomputer independently controls each LED lamp core through pulse width.

[0053] The single chip microcomputer adjusts the duty cycle of the output signal by changing the pulse width of the control signal.

[0054] The control signal consists of a periodic square wave with a fixed frequency and a rectangular pulse with a variable duty cycle; the high level of the square wave represents the signal "1", and the low level represents the signal "0"; the width of the rectangular pulse represents the duration of the high level, and the duty cycle is determined by the proportion of the duration of the high level to the entire period; the brightness of the LED is proportional to the intensity of the current passing through it; through the PWM technology, the current intensity of the LED is accurately controlled, thereby achieving precise control of the brightness.

[0055] Industrial cameras are only responsible for shooting, while the corner points are processed and analyzed by the camera's host computer. The host computer uses machine vision algorithms to identify uploaded images.

[0056] The preferred embodiments of the present invention disclosed above are only used to help illustrate the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to the specific implementation methods described. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for visual inspection of curved surfaces, characterized in that: include: Choose any point p in the image, the brightness value of the pixel at point p is Ip; Judging whether point p is the center point of a corner point includes: Set a threshold t. The threshold t compares the grayscale value difference between point p and its surrounding pixels. A circle is constructed with point p as the center and r as the radius, covering M pixels around point p. If r=3, then M=16, that is, there are 16 pixels on the circumference. The grayscale values ​​of the 16 pixels on the circumference are compared to obtain pixels whose grayscale values ​​exceed Ip+t and are lower than Ip-t. If there are n consecutive pixels that satisfy the condition that they are all greater than Ip+t or less than Ip-t, then point p is considered to be a corner point.

2. The method according to claim 1, characterized in that Fast detection of corner points in an image includes: Detect the grayscale values ​​of the four pixels numbered 1, 9, 5, and 13 on the circumference. If 3 or more pixels among these four points meet the condition that they are all greater than Ip+t or less than Ip-t, then the point p is a corner point, otherwise it is directly excluded.

3. The method according to claim 2, characterized in that Avoiding corner point concentration and performing non-maximum suppression include: Retain the corner points with the maximum response value in a certain area; The corner point with the maximum response value is obtained by calculating the maximum value of the minimum value of all consecutive n pixels in the absolute difference between the corner point and the 16 pixels on its circumference; If point p satisfies the grayscale values ​​of n consecutive pixels and is still retained after non-maximum suppression, then the point is determined to be a corner point.

4. A curved surface visual inspection method according to claim 3, characterized in that: It also includes a DNN module, which imports and runs models trained under different deep learning frameworks, and uses the FAST feature extraction algorithm by combining the OpenCV with the deep learning framework; Comparing the grayscale values ​​of a pixel with its surrounding neighborhood pixels is achieved through the FAST feature extraction algorithm.

5. An industrial camera for arc surface visual inspection, based on the method according to any one of claims 1 to 4, characterized in that: The industrial camera body (2) is provided with a camera lens (3), a ring-shaped arc-shaped concave light source (1) is provided on one side of the industrial camera body (2), the ring-shaped arc-shaped concave light source (1) is detachable, 16 LED lamp cores are evenly arranged on the concave surface of the ring-shaped arc-shaped concave light source (1), and a single-chip microcomputer is provided inside the industrial camera body (2), and the single-chip microcomputer independently controls each LED lamp core through pulse width.

6. The arc surface visual inspection industrial camera according to claim 6, characterized in that: The single chip microcomputer adjusts the duty cycle of the output signal by changing the pulse width of the control signal.

7. The arc surface visual inspection industrial camera according to claim 7, characterized in that: The control signal consists of a periodic square wave with a fixed frequency and a rectangular pulse with a variable duty cycle; the high level of the square wave represents signal "1", and the low level represents signal "0"; the width of the rectangular pulse represents the duration of the high level, and the duty cycle is determined by the proportion of the duration of the high level to the entire period; the brightness of the LED is proportional to the intensity of the current passing through it; through PWM technology, the current intensity of the LED is accurately controlled, thereby achieving accurate control of the brightness.

Images