CN110223339A - One kind being based on machine vision thermal protector calibration point center positioning method - Google Patents

Abstract

The invention discloses a method for locating the center of the calibration point of a thermal protector based on machine vision, including image preprocessing, target extraction, edge extraction of the calibration point area, and determination of the center position of the calibration point. The invention uses an improved random Hough transform to extract The edge of the circle is detected to obtain the center position of the calibration point. The present invention overcomes the situation that the traditional Hough transform detection circle has a large amount of calculation and takes up too much memory, and greatly improves the detection speed; the present invention overcomes the problem that the random circle detection algorithm (RCD) detects the edge points of the image in complex backgrounds. There is a problem of poor robustness, and the robustness of the system is improved, so that the thermal protector calibration point center positioning can enhance the adaptability to the environment, and the application field of the system is wider.

Description

A method for center positioning of calibration points of thermal protector based on machine vision

technical field

The invention belongs to the technical field of thermal protectors, and in particular relates to a method for locating the center of a calibration point of a thermal protector based on machine vision.

Background technique

Thermal protector is a kind of circuit protection device, which is widely used in various electrical products. Its main working principle is to keep the circuit connected at room temperature, and cut off the circuit when the temperature exceeds the set threshold. The main function is the internal bimetal, but the jump temperature of the bimetal is closely related to the depth of the calibration point, so the depth measurement of the calibration point is an important content in product inspection. Due to the small size of the calibration point, the first step when using structured light depth detection is to locate the center point of the calibration point to ensure that the structured light can accurately hit the deepest part of the calibration point.

The two-dimensional imaging geometric feature of the thermal protector calibration point is a circular structure, so the center of the calibration point can be located by first detecting the circle and then obtaining the coordinates of the center of the circle. For circle detection, there are currently a large number of related research and detection methods. The Hough transform detection circle is the most basic detection method. It transforms the plane coordinates into parameter coordinates, accumulates the coordinate points in the parameter space, and obtains the peak point to obtain the circle parameters. The traditional Hough change detection circle is computationally intensive and takes up too much memory. Aiming at the deficiencies of traditional detection methods, scholars at home and abroad have proposed a variety of improved methods on this basis. (1) Xu et al. proposed the Random Hough Transform (RHT), which alleviated the problems of large computation and large memory usage to a certain extent. RHT randomly samples three non-collinear edge points in the image space and maps them to the parameter space, and extracts the circle parameters corresponding to the peak values. This many-to-one mapping greatly improves the detection speed; although the RHT algorithm is in the plane coordinates to The many-to-one method is used for parameter coordinate mapping, but there is still the problem of spatial parameter accumulation, so the calculation speed needs to be improved. (2) The random circle detection algorithm (RCD) proposed by Chen et al., RCD directly detects the edge points of the image, there is no problem of spatial parameter accumulation, and the detection speed and detection accuracy are improved, but this method is complex There is a problem of poor robustness in the background. (3) Li Fuqing and Su Zhan proposed a circle detection method based on particle swarm optimization algorithm. Compared with the Hough transform circle detection algorithm, the particle swarm detection algorithm has higher efficiency and shorter running time.

Contents of the invention

In order to solve the problems of complex optical conditions, many interference factors and unsatisfactory detection effect of current thermal protectors, the present invention proposes a center positioning method of calibration points of thermal protectors based on machine vision. This method is improved on the original circle detection method, and an improved Hough transform circle detection algorithm is proposed to realize the detection and center location of the optimal circle at the edge of the calibration point.

Technical scheme of the present invention is as follows:

A method for locating the center of a calibration point of a thermal protector based on machine vision, comprising the following steps:

The first step is image preprocessing, which specifically includes: using an industrial camera with a microscope lens to collect images of thermal protectors, grayscale processing of color images, filtering of images, and threshold segmentation of images;

The second step, target extraction, uses the open operation in morphology to remove character areas, smooths the image contour, disconnects narrow connections and eliminates burrs, and uses the connected domain area constraint algorithm to eliminate the edge formation after morphological processing. The discrete points of , by marking the connected domain, determine the calibration point area;

The third step is to extract the edge of the calibration point area;

The fourth step is to determine the center position of the calibration point, and perform circle detection on the extracted edge through the improved random Hough transform to obtain the center position of the calibration point.

Specifically, in the third step of calibration point region edge extraction, the 8-neighborhood region growing algorithm is used to process the binarized edge image to determine connected regions.

The 8-neighborhood region growing algorithm specifically includes the following steps to process the binarized edge image:

Step 1, scan the image sequentially from top to bottom and from left to right, and when the first point with a pixel of 1 is scanned, use this point as the base point and mark the point;

Step 2, take the base point as the seed point, and mark the target pixels in its 8 neighborhoods in the same way;

Step 3, mark the target pixels in the 8 neighborhoods of all marked pixels with the same label until the connected region is marked;

Step 4, continue to scan sequentially, and repeat the first three steps, until all point markers with a pixel value of 1 end.

Specifically, the determination of the center position of the calibration point in the fourth step specifically includes:

Step 1, load the image;

Step 2, scan the image from left to right and from top to bottom to obtain a point on the edge, find the corresponding point in the vertical direction of the edge subset, and find the center of symmetry of the two points;

Step 3, rotate 900 around the center of symmetry with the straight line where the two points are located;

Step 4, obtain the two intersection points of the straight line and the edge after rotation, use these two intersection points as points on the circle, and use the central symmetry point of the two intersection points as the center of the circle to determine a circle equation C1;

Step 5: Take the diameter of the circle as the axis and the center of the circle as the center of rotation, and rotate 30 degrees clockwise to get two new points:

s=(x-a)cos30°-(y-b)sin30°+a

t=(y-b)cos30°-(x-a)sin30°+b

Among them, x, y represent the intersection point of the diameter extension line and the edge, a, b are the symmetrical center coordinates of the two intersection points, s, t represent the coordinates after rotation, obtain the coordinates after rotation (s0, y0), (s1, t1); find A straight line passing through two points of intersection:

Comparing the straight line and the edge with two points, find the central symmetry point, determine a circle C1, and repeat this step until the cumulative sum of the rotation angle exceeds 180°;

Step 6, set the number of edge pixels on the i-th circle as E(S _i ), point f(s _i , t _i ) is the pixel value of the point on the circle, then E(S _i ) can be expressed as:

Use N to represent the number of edge point pixels, then the deviation coefficient of each circle is:

Step 7, the circle with the smallest deviation value coefficient is the optimal circle, and the point where the center of the optimal circle is located is the center position of the calibration point.

Compared with prior art, the beneficial effect of the present invention is:

(1) It overcomes the traditional Hough transform detection circle, which has a large amount of calculation and takes up too much memory, and greatly improves the detection speed.

(2) To overcome the problem of poor robustness in the complex background when the random circle detection algorithm (RCD) detects the edge points of the image, and improve the robustness of the system.

(3) The positioning accuracy of the center position of the calibration point is improved, and the relative error is reduced.

(4) The central positioning of the calibration point of the thermal protector enhances the adaptability to the environment, and the application field of the system is wider.

Description of drawings

Fig. 1 is a flow chart of the method of the present invention;

Fig. 2 is a schematic diagram of a rotation candidate circle in the present invention;

Fig. 3 is a schematic diagram of non-unique highest point error of the present invention;

Fig. 4 is a flow chart of determining the optimal circle center in the present invention.

Detailed ways

The technical solutions in the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the present invention.

Figure 1 is a flow chart of the calibration point center positioning process in the present invention, including image preprocessing, target extraction, calibration point area edge extraction, and calibration point center position determination.

The image preprocessing includes the following steps: step 1, the diameter of the calibration point is about 1 mm, which belongs to the category of micro-size measurement, in order to obtain a clear image of the calibration point, use an industrial camera with a micro lens to collect images of the thermal protector; step 2, In order to reduce the amount of calculation and improve the calculation speed, grayscale processing is performed on the color image; step 3 is to eliminate the influence of metal reflection and external environment, and filter the image; step 4 is to perform threshold segmentation on the image.

The preprocessed image contains a circular structure of numbers and English letters, which is composed of double edges, and the edge width is between 4 and 8 pixels, while the edge width of the calibration point area is more than 10 pixels, so the target extraction is to use The opening operation in the morphology performs de-character area, smoothes the image contour, breaks the narrow connection and eliminates the glitch. After the morphological processing, there will be discrete points on the edge. By using the connected domain area constraint algorithm, the discrete points will be removed, and the calibration point area will be determined by marking the connected domain.

The edge extraction of the calibration point area is to use the 8-neighborhood area growing algorithm to process the binarized edge image to determine the connected area. The method comprises the following steps: step 1, scan the image sequentially from top to bottom and from left to right, and when the first pixel is 1 is scanned, use this point as a base point and mark the point; step 2, Using the base point as the seed point, mark the target pixels in the 8 neighborhoods with the same label; step 3, mark the target pixels in the 8 neighborhoods of all marked pixels with the same label until the connected region is marked; step 4 , continue to scan sequentially, and repeat the first three steps until the end of all point markers with pixel values of 1.

The determination of the center position of the calibration point is to perform circle detection on the extracted edge through the improved random Hough circle detection algorithm to obtain the center position of the calibration point. The process is shown in Figure 4:

As shown in Figures 2 and 3, the circle is a centrally symmetrical figure. First, scan from left to right and from top to bottom to get a point on the edge. Find the corresponding point in the vertical direction of the edge subset, and find the symmetry center of the two points. Rotate the straight line where the two points are located by 900 around the center of symmetry. After the rotation, the two intersections of the straight line and the edge are (x ₀ , y ₀ ), (x ₁ , y ₁ ), respectively. The two intersections are taken as points on the circle, and the two intersections The central symmetry point of is taken as the center of the circle, and a circle C1 is determined. Because the inner edge is not a completely regular circle, the first edge point obtained by the first scan is not necessarily the only highest point of the circle. When the edge point is not the only highest point, the line l1 connecting it with the symmetric point is where The chord of the circle can eliminate the error caused by the non-unique highest point by rotating 90°, that is, the diameter of the circle is perpendicular to the chord. As shown in Figure 2, the intersection of l2 and the edge can determine a circle. Taking the diameter of the circle as the axis and the center of the circle as the center of rotation, rotate 30 degrees clockwise to get two new points:

s=(x-a)cos30°-(y-b)sin30°+a

t=(y-b)cos30°-(x-a)sin30°+b

Where x, y represent the intersection of the diameter extension line and the edge, a, b are the symmetrical center coordinates of the two intersections, s, t represent the coordinates after rotation. Obtain the rotated coordinates (s0, y0), (s1, t1) to find the straight line where the two intersection points are located:

The straight line and the edge intersect at two points, find the central symmetry point, determine a circle C2, and repeat this step until the cumulative sum of the rotation angle exceeds 180°. In Figure 3, C1, C2, and C3 are circles corresponding to rotations of 0°, 30°, and 90°, respectively.

Let the number of edge pixels on the i-th circle be E(S _i ), and the point f(s _i , t _i ) be the pixel value of the point on the circle, then E(S _i ) can be expressed as:

use N

The circle with the smallest deviation value coefficient is the optimal circle. The point where the center of the optimal circle is located is the center position of the calibration point, and its flow chart is shown in Figure 4.

Claims (4)

1. A kind of center location method based on machine vision thermal protector calibration point, it is characterized in that: comprise the following steps: The first step is image preprocessing, which specifically includes: using an industrial camera with a microscope lens to collect images of thermal protectors, grayscale processing of color images, filtering of images, and threshold segmentation of images; The second step, target extraction, uses the open operation in morphology to remove character areas, smooths the image contour, disconnects narrow connections and eliminates burrs, and uses the connected domain area constraint algorithm to eliminate the edge formation after morphological processing. The discrete points of , by marking the connected domain, determine the calibration point area; The third step is to extract the edge of the calibration point area; The fourth step is to determine the center position of the calibration point, and perform circle detection on the extracted edge through the improved random Hough transform to obtain the center position of the calibration point. 2. A kind of center positioning method based on machine vision thermal protector calibration point according to claim 1, it is characterized in that: in the third step, the 8-neighborhood area growth algorithm is used to process the binarized edge image, and determine the connectivity area. 3. A kind of center positioning method based on machine vision thermal protector calibration point according to claim 2, it is characterized in that: specifically comprise the following steps: Step 1, scan the image sequentially from top to bottom and from left to right, and when the first point with a pixel of 1 is scanned, use this point as the base point and mark the point; Step 2, take the base point as the seed point, and mark the target pixels in its 8 neighborhoods in the same way; Step 3, mark the target pixels in the 8 neighborhoods of all marked pixels with the same label until the connected region is marked; Step 4, continue to scan sequentially, and repeat the first three steps, until all point markers with a pixel value of 1 end. 4. A method for locating the center of the calibration point of a thermal protector based on machine vision according to claim 1, characterized in that: the fourth step specifically includes: Step 1, load the image; Step 2, scan the image from left to right and from top to bottom to obtain a point on the edge, find the corresponding point in the vertical direction of the edge subset, and find the center of symmetry of the two points; Step 3, rotate 90° around the center of symmetry with the straight line where the two points are located; Step 4, obtain the two intersection points of the straight line and the edge after rotation, use these two intersection points as points on the circle, and use the central symmetry point of the two intersection points as the center of the circle to determine a circle equation C1; Step 5: Take the diameter of the circle as the axis and the center of the circle as the center of rotation, and rotate 30 degrees clockwise to get two new points: s=(x-a)cos30°-(y-b)sin30°+a t=(y-b)cos30°-(x-a)sin30°+b Among them, x, y represent the intersection point of the diameter extension line and the edge, a, b are the symmetrical center coordinates of the two intersection points, s, t represent the coordinates after rotation, obtain the coordinates after rotation (s0, y0), (s1, t1); find A straight line passing through two points of intersection: Comparing the straight line and the edge with two points, find the central symmetry point, determine a circle C1, and repeat this step until the cumulative sum of the rotation angle exceeds 180°; Step 6, set the number of edge pixels on the i-th circle as E(S _i ), point f(s _i , t _i ) is the pixel value of the point on the circle, then E(S _i ) can be expressed as: Use N to represent the number of edge point pixels, then the deviation coefficient of each circle is: Step 7, the circle with the smallest deviation value coefficient is the optimal circle, and the point where the center of the optimal circle is located is the center position of the calibration point.