CN111815575A - A detection method of bearing steel ball parts based on machine vision - Google Patents

Abstract

A bearing steel ball part detection method based on machine vision comprises steel ball circularity detection and part scratch detection and comprises the following steps: the method comprises the following steps of firstly, detecting the circularity of a steel ball, and the process is as follows: 1.1) preprocessing an input image; 1.2) acquiring the outline of the steel ball; 1.3) screening the obtained edges, and outputting the outline and circle center radius of the steel ball qualified by detection; step two, detecting scratches of the parts: 2.1) detecting the edge of the front surface of the part; 2.2) detecting the side edge of the part. The invention provides a bearing steel ball part detection method based on machine vision, which can overcome various interference factors and establish a high-efficiency and stable detection system.

Description

A detection method of bearing steel ball parts based on machine vision

technical field

The invention relates to the field of parts detection, in particular to a method for detecting bearing steel ball parts based on machine vision.

Background technique

While promoting social progress, human beings hand over many complex and repetitive tasks to machines, and machine vision systems fall into this category. Generally speaking, a machine vision system is an industrial test monitoring system. Machine vision systems are often used to improve the quality or automation of production lines in dangerous workplaces or situations that are difficult to recognize by the human eye. Machine vision technology is a direction in the field of detection. Its purpose is to replace human eyes with machines to complete tasks such as detection, recognition, and classification. Complete machine vision technology includes light source lighting, optical imaging, digital image processing, machine classification, industrial control and other technologies, with high stability and accuracy. In foreign countries, machine vision is mainly used in semiconductor, electronics and machinery industries, such as integrated circuit manufacturing, electronic molds, screen printing, component molding, etc. In addition, machine vision is also widely used in product quality inspection systems and plays an important role in industrial production.

Existing parts inspection methods are generally based on machine vision technology. The researchers developed a set of circular parts detection framework based on machine vision technology, and realized the precise measurement of circular parts through simple camera calibration and sub-pixels. The algorithm model mainly includes system correction, preprocessing, image binarization and circular Shape detection in four parts. Others have analyzed and studied the precision machining technology of aspherical curved optical parts, and developed a set of ultra-precision air compression spindle system, which improves the detection accuracy to the international level. A line scan industrial camera is used to scan the image of the computer hard disk thin part to be detected, and a new calibration algorithm and contour vectorization algorithm are proposed according to the characteristics of image scanning. Get the size parameters of the parts to be inspected. Some researchers detect the defects on the surface of the part, and accurately extract the detection area of the target part through two dynamic extraction algorithms, and then generate detection parameters based on the statistical map classification algorithm and the edge information such as the width, thickness, dispersion, and deflection of the part. . In modern automated assembly lines, the inspection system can confirm the size, quantity, and presence of defects of parts anytime and anywhere, which greatly improves industrial production efficiency. However, the problems of current machine vision are also obvious, and problems such as system time delay, low precision, and poor stability need to be solved urgently.

SUMMARY OF THE INVENTION

In order to overcome the inability to detect parts quickly and accurately in the prior art, and to solve the above technical problems, the present invention provides a method for detecting bearing steel ball parts based on machine vision, which can overcome various interference factors and establish an efficient and stable detection system .

The technical scheme adopted by the present invention to solve its technical problems is:

A machine vision-based detection method for bearing steel ball parts, characterized in that the detection method comprises the following steps:

The first step is to test the roundness of the steel ball. The process is as follows:

1.1) Image preprocessing: use Gaussian filtering to preprocess the input image to remove high-frequency signals in the image;

1.2) Extracting the ostium of the superior mesenteric artery: The image is transformed by grayscale conversion, threshold segmentation and other methods to make it easier for subsequent algorithms to identify image information; after this step, the system will obtain several closed edge contours;

1.3) Screening edge contours: In order to obtain better detection edges, three screening conditions are adopted: contour pixel domain detection, ellipse fitting detection and circularity detection. Contour pixel domain detection is to count the pixel area of each contour; Since the circular contour of the steel ball will form an ellipse similar to a circle after imaging, the ellipse fitting can effectively eliminate noise interference; finally, after obtaining the edge contour information of the steel ball, the circularity of the ellipse is detected. The purpose of this step is to detect whether the shape of the steel ball conforms to the production specification;

The second step, parts scratch detection, the process is as follows:

2.1) Scratch detection on the side of the part: use the edge detection method to detect the scratch on the side of the part, first determine the edge pixels in the image, and then connect these pixels together to form the required area boundary;

2.2) Scratch detection on the front of the part: Because there is too much edge information on the front, it will have a serious impact on the classification results. The method of eliminating redundant edge information is used to remove high-frequency signals in the image.

Further, in 2.1), the edge of the image represents the beginning of one area and another area in the image, and the set of adjacent pixels in the image constitutes the edge of the image, so the edge of the image can be understood as the grayscale of the image The collection of pixels with spatial mutation, the image edge has two important concepts: direction and gradient. The pixels along the edge direction change smoothly, while the pixels perpendicular to the edge direction change more violently. Therefore, according to this feature, it is usually used. The first-order and second-order derivatives are used for edge detection. Therefore, the edge detection in an image can be determined by derivation of the gray value, and the operation of the derivative can be performed by a differential operator.

Further, the steps of described 2.2) are as follows:

2.2.1) Preprocess the image with smooth filtering to remove high-frequency signals in the image;

2.2.2) Use edge detection method to detect image edge information and extract the closed contour of the image;

2.2.3) use ellipse fitting method to carry out ellipse fitting to outline, and the reason why ellipse fitting is to be carried out is because the target area to be extracted by the present invention is a ring;

2.2.4) According to the fixed information of the image, select an appropriate threshold value, screen the obtained elliptical outline, and obtain the inner circle and outer circle outline of the ring area, and finally use the method of image subtraction to obtain the target area to be extracted. .

The beneficial effects of the invention are mainly manifested in: based on machine vision, the problem that it is difficult to manually identify defective products in the industrial production process is solved. And the method has high stability and transplantability, and has broad application prospects.

Description of drawings

Fig. 1 is the machine vision detection system of the present invention;

Fig. 2 is the circularity detection schematic diagram of the present invention;

FIG. 3 is the flow of the detection of the front side of the part of the present invention.

Detailed ways

The present invention will be further described below in conjunction with the accompanying drawings.

Referring to Figures 1 to 3, a method for detecting bearing steel ball parts based on machine vision mainly solves the detection of the roundness of steel balls and the detection of parts scratches, including the following steps:

The first step is to detect the circularity of steel balls. The process of the detection method is shown in Figure 2. The process is as follows:

1.1) Preprocess the input image, the specific steps are as follows:

The present invention first uses Gaussian filtering to preprocess the input image to remove high-frequency signals in the image;

1.2) To obtain the steel ball profile, the specific steps are as follows:

After obtaining the pre-processed information, the present invention transforms the image through grayscale conversion, threshold segmentation and other methods, so that the subsequent algorithm can more easily identify the image information. After this step, the system will obtain several closed edge contours, only one of which is the edge contour of the steel ball, and the other contours are due to noise interference generated by lighting;

1.3) Screen the obtained edge contour, and output the contour and center radius of the steel ball that has passed the inspection. The steps are as follows:

In order to obtain a better detection edge, the present invention adopts three screening conditions: contour pixel domain detection, ellipse fitting detection and circularity detection. Contour pixel domain detection is to count the pixel area of each contour. Since the circular outline of the steel ball will form an ellipse similar to a circle after imaging, ellipse fitting can effectively eliminate noise interference. Finally, after the edge contour information of the steel ball is obtained, the circularity of the ellipse is detected. The purpose of this step is to detect whether the shape of the steel ball conforms to the production specification. After the system imaging, the circularity of the contour should be within a certain error range. , steel balls beyond this range are likely to be defective parts that need to be removed. The pixel thresholds in this article are: 1600, 2000, and the circularity threshold is 0.980;

Circularity is used to describe the complexity of the image target boundary, and its value is the smallest when the target is a perfect circle shape. The most commonly used circularity is the ratio of the square of the circumference of the circumference to the area. The more complex the shape, the greater the ratio. The circularity index achieves a minimum value of 4π when it is a perfect circle. Take C as the circularity index, P as the perimeter, and A as the area. The calculation formula is as follows:

The above calculation method is relatively rough, and the present invention uses boundary energy to measure. Take p as the distance from any point on the boundary to a starting point. At any point, the boundary has an instantaneous radius of curvature r(p), which can be known from the geometric relationship as the radius of the circle tangent to the point and the boundary. The curvature function at point p is as follows:

The function K(p) is a periodic function with period p. The formula for calculating the average energy per unit boundary length is as follows:

Take the radius of the circle as R. The curvature can be calculated from the chain code, so the boundary energy can also be easily obtained. For a boundary with a fixed area, the minimum boundary energy E ₀ of a circle is:

where p is the distance from any point on the boundary to a starting point, and R is the radius of the circle;

The second step, parts scratch detection, the process is as follows:

2.1) Scratch detection on the side of the part, the basic steps are as follows:

The invention uses the edge detection method to detect the scratches on the side of the part, firstly determines the edge pixels in the image, and then connects these pixels together to form the required area boundary. The edge of an image represents the beginning of one area and another area in the image, and the collection of adjacent pixels in the image constitutes the edge of the image. Therefore, the edge of an image can be understood as a collection of pixels where the grayscale of the image undergoes spatial abrupt changes. Image edges have two important concepts: direction and gradient. The pixel changes along the edge direction are relatively smooth, while the pixel changes perpendicular to the edge direction are more severe. Therefore, according to this feature, the first and second derivatives are usually used for edge detection. Therefore, the edge detection in an image can be determined by derivation of the gray value, and the operation of the derivative can be performed by the differential operator;

The Sobel operator is a first-order differential operator, which calculates the gradient of each pixel through the gradient value of the adjacent area of the pixel, and finally makes a choice according to a fixed threshold. Its calculation formula is as follows:

The Sobel operator is a three-layer operator template. The Sobel operator is formed by two convolution kernels of dx and dy. One of the convolution kernels performs the calculation of the vertical edge, one of the convolution kernels performs the calculation of the horizontal convolution kernel, and the maximum value of the two convolution kernels is used as the final output result of the calculation.

In addition to the Sobel operator, this paper also uses the Canny operator for edge detection. In this paper, the Canny operator is used to calculate the local maximum value of the image gradient, and two thresholds are used to find the strong and weak edges of the target. The method is essentially smoothing through a quasi-Gaussian function, and then locating the maximum value of the derivative through a first-order differential operator. Within the second-order square, the mean of the finite differences can be computed to find the gradient of the partial derivative at a point in the image. The direction angle and magnitude can be calculated by the coordinate conversion formula from the rectangular coordinate system to the polar coordinate system:

where M[i,j] reflects the edge strength of the image, and θ[i,j] reflects the edge direction of the image. When M[i,j] obtains a local maximum value, θ[i,j] is the edge direction at this time. For the gradient amplitude, the non-maximum suppression method is used to extract the pixel with the largest gradient value in each gradient direction;

2.2) The front edge detection of parts, the detection process is shown in Figure 3, because too much front edge information will have a serious impact on the classification results, so the present invention proposes a method for eliminating redundant edge information. , the steps are as follows:

2.2.1) Preprocess the image with smooth filtering to remove high-frequency signals in the image;

2.2.2) Use edge detection method to detect image edge information and extract the closed contour of the image;

2.2.3) use ellipse fitting method to carry out ellipse fitting to outline, and the reason why ellipse fitting is to be carried out is because the target area to be extracted by the present invention is a ring;

2.2.4) According to the fixed information of the image, select an appropriate threshold value, screen the obtained elliptical outline, and obtain the inner circle and outer circle outline of the ring area, and finally use the method of image subtraction to obtain the target area to be extracted. .

The above descriptions are only the embodiments of the present invention, and are not intended to limit the scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the description and drawings of the present invention, or directly or indirectly applied to other related The technical field of the present invention is similarly included in the scope of patent protection of the present invention.

Claims (3)

1. a bearing steel ball parts detection method based on machine vision, is characterized in that, described detection method comprises the following steps: The first step is to test the roundness of the steel ball. The process is as follows: 1.1) Image preprocessing: use Gaussian filtering to preprocess the input image to remove high-frequency signals in the image; 1.2) Obtain the outline of the steel ball: transform the image through grayscale conversion, threshold segmentation and other methods, so that the subsequent algorithm can more easily identify the image information; after this step, the system will obtain several closed edge contours; 1.3) Screening edge contours: In order to obtain better detection edges, three screening conditions are adopted: contour pixel domain detection, ellipse fitting detection and circularity detection. Contour pixel domain detection is to count the pixel area of each contour; Since the circular contour of the steel ball will form an ellipse similar to a circle after imaging, the ellipse fitting can effectively eliminate noise interference; finally, after obtaining the edge contour information of the steel ball, the circularity of the ellipse is detected. The purpose of this step is to detect whether the shape of the steel ball conforms to the production specification; The second step, parts scratch detection, the process is as follows: 2.1) Scratch detection on the side of the part: use the edge detection method to detect the scratch on the side of the part, first determine the edge pixels in the image, and then connect these pixels together to form the required area boundary; 2.2) Scratch detection on the front of the part: Because there is too much edge information on the front, it will have a serious impact on the classification results. The method of eliminating redundant edge information is used to remove high-frequency signals in the image. 2. The method for detecting bearing steel ball parts based on machine vision as claimed in claim 1, characterized in that, in 2.1), the edge of the image represents the beginning of one area and another area in the image, and adjacent areas in the image. The set of pixels in between constitutes the edge of the image. Therefore, the edge of the image can be understood as the set of pixels whose grayscale of the image undergoes spatial mutation. There are two important concepts in the image edge: direction and gradient, along the direction of the edge. The change of the pixels of the image is relatively smooth, while the change of the pixels perpendicular to the edge direction is more severe. Therefore, according to this characteristic, the first and second derivatives are usually used for edge detection. Therefore, the edge detection in an image can be detected by grayscale. The degree value is derived to determine it, and the operation of the derivative can be carried out by the differential operator. 3. a kind of bearing steel ball parts detection method based on machine vision as claimed in claim 1 or 2, is characterized in that, the step of described 2.2) is as follows: 2.2.1) Preprocess the image with smooth filtering to remove high-frequency signals in the image; 2.2.2) Use edge detection method to detect image edge information and extract the closed contour of the image; 2.2.3) use ellipse fitting method to carry out ellipse fitting to outline, and the reason why ellipse fitting is to be carried out is because the target area to be extracted by the present invention is a ring; 2.2.4) According to the fixed information of the image, select an appropriate threshold value, screen the obtained elliptical outline, and obtain the inner circle and outer circle outline of the ring area, and finally use the method of image subtraction to obtain the target area to be extracted. .

Images