CN110308153A - Metal workpiece defect detection method, system, storage medium, and device based on monocular stereo vision - Google Patents

Abstract

The invention discloses a metal workpiece defect detection method based on monocular stereo vision, which is used for defect inspection of the metal workpiece to be tested, comprising the following steps: Step S1: centering on the metal workpiece to be tested, setting Light source and perform light source direction calibration to obtain the final light source direction vector in each orientation; and step S2: preset template workpiece, and obtain images of the metal workpiece to be tested and the template workpiece in at least three orientations under the final light source direction vector ; Use the principle of photometric stereo vision to digitize the image and process it for comparative analysis to determine whether there is a defect in the metal workpiece to be tested. Based on the above method, the present invention also provides a system; the present invention also provides a storage medium; and the present invention also provides a device.

Description

Metal workpiece defect detection method, system and storage medium based on monocular stereo vision quality, and device

【Technical field】

The invention relates to the field of intelligent supplementary light, in particular to a metal workpiece defect detection method, system, storage medium, and device based on monocular stereo vision.

【Background technique】

Metal workpieces are the basic components in daily industrial production, mainly referring to metal blocks, metal rods, metal pipes, etc. of various specifications and shapes made of metal materials, but the quality of metal workpieces has always been attached great importance to in our country. At present, most domestic manufacturers still use manual visual inspection to detect steel balls, and a large number of inspectors use visual methods to detect steel balls under incandescent lamps and perform simple sorting on them. The detection method relies on the experience and sense of responsibility of the inspectors. It is labor-intensive and low-efficiency. Repeated operations for a long time can easily cause visual fatigue, and the detection results are random. Therefore, the technology of using machine vision to detect defects in metal workpieces emerged as the times require. ;

In today's technology, the breakthroughs faced by machine vision detection of metal workpiece defects are nothing more than two points: the first point is how to obtain a full range of high-definition images of metal workpieces because surface defects are randomly distributed on metal workpieces; second Due to the characteristics of the metal or the effect of post-processing and polishing, the surface reflectance of the metal workpiece is quite high, resulting in uneven brightness distribution when collecting images, which often submerges the defect information to be detected. For example, when collecting images of metal workpieces, by There are often some too bright light spots or too dark black blocks, which makes the defect information inaccurate during the detection process, so how to find a suitable light source direction for image acquisition;

The above two problems are common in the application of metal workpieces, and image recognition technology focuses on and needs to be solved urgently.

【Content of invention】

In order to overcome the above problems, the present invention provides a metal workpiece defect detection method based on monocular stereo vision.

The present invention provides a technical solution as follows: The present invention provides a metal workpiece defect detection method based on monocular stereo vision, which is used for defect inspection of the metal workpiece to be tested, including the following steps: Step S1: centering on the metal workpiece to be tested, Calibrate the direction of the light source in at least three directions to obtain the final direction vector of the light source in each direction; and step S2: preset the template workpiece, and obtain the metal workpiece to be tested and the template workpiece under the final light source direction vector in at least three directions The image; the principle of photometric stereo vision is used to digitize the image and process it for comparative analysis to determine whether there is a defect in the metal workpiece to be tested.

Preferably, in step S1, centering on the metal workpiece to be tested, the direction of the light source is calibrated in at least three directions to obtain the final direction vector of the light source in each direction, which specifically includes the following steps: Step S11: in at least three directions Set the light source in each orientation; thus obtain the light source in at least three orientations, introduce the calibration ball and place it at equal intervals five times in the light source irradiation range of each orientation, and carry out the calibration ball under the light source in all orientations. Acquisition, thus at least 15 pairs of calibration ball images are obtained; step S12: identify the center point and highlight point of the calibration ball in each calibration ball image, and calculate the spatial position relationship between the center point and the highlight point; regard the calibration ball as orthogonal The projection model is combined with the principle of photometric stereo vision to calculate the light source direction vector in each orientation; and step S13: use the least squares method to obtain the final light source in each orientation through the light source direction vector obtained from each calibration ball image direction vector.

Preferably, in step S2, the template workpiece is preset, and the images of the metal workpiece to be tested and the template workpiece in at least three orientations under the final light source direction vector are obtained; the image is digitized and processed for comparison and analysis by using the principle of photometric stereo vision to Judging whether there is a defect in the metal workpiece to be tested specifically includes the following steps: Step S21: Obtain at least 3 images of the metal workpiece to be tested and the preset template workpiece under at least three final light source direction vectors; Step S22: Obtain according to step S21 Using the Phong model based on the orthogonal projection model, the surface normal vector of the metal workpiece to be tested is obtained and the gradient map is respectively established, and the gradient map is Gaussian transformed to obtain the curvature map, and the curvature map is morphologically processed; and Step S23: Bolb analysis is performed on the morphologically processed curvature map of the metal workpiece to be tested, so as to output whether there is a defect in the metal workpiece to be tested.

Preferably, step S3 is further included, specifically including the following steps:

Step S3: According to the result of step S2, the metal workpieces to be tested are sorted and grasped by the mechanical arm.

In order to better solve the above problems, the present invention proposes another solution as follows, a metal workpiece defect detection system based on monocular stereo vision, which is used for defect inspection of the metal workpiece to be tested, including:

Emitting module: used to emit light from at least three directions of the metal workpiece to be tested;

Acquisition module: used to collect the image of the metal workpiece to be tested under the irradiation of the emission module;

Processing module: used to calculate the final light source direction vector of the emission module to the metal workpiece in each of the eight orientations, and to determine whether the metal workpiece to be tested has defects;

Execution module: classify the metal workpiece to be tested;

Control module: according to the result of the processing module, it is used to control the launch module and the execution module to perform corresponding actions.

In order to better solve the above problems, the present invention proposes another solution as follows, a storage medium, the storage medium or the processor stores a computer program, and when the storage medium program is running, the computer program to be executed by the storage medium or the processor is controlled It is necessary to implement the metal workpiece defect detection method based on the above-mentioned monocular stereo vision.

In order to better solve the above problems, the present invention proposes another solution as follows, a metal workpiece defect detection device based on monocular stereo vision, which is used for defect inspection of the metal workpiece to be tested, which includes a camera, a light source, a circular light source controller, stage, industrial computer and computer; the industrial computer is electrically connected to the camera, light source, circular light source controller, stage, and computer, and the light source and the camera are located at the On the side of the circular light source controller facing the stage, and the camera is located at the center of the circular light source controller, the metal component to be tested can be placed on the stage, wherein the light source is based on the camera The center is evenly distributed in a ring shape, and the light sources are at least three, so that light is emitted in at least three directions so that the light completely covers the metal workpiece to be tested; when the metal workpiece to be tested is placed on the stage, the industrial control The camera controls the circular light source controller to start responding, and the circular light source controller can control each of the light sources to be turned on individually; when each of the light sources is individually turned on, the camera collects an image and transmits it to the The computer analyzes to find the final light source direction vector in each orientation; wherein, when the computer analyzes the image, the method of step S1 in claim 1 needs to be used.

Preferably, the device further includes a driving rail and a mechanical arm, the object stage is arranged on the driving rail and is fixedly connected, and the industrial computer can control the operation of the driving rail so that the object stage is on the driving rail It can move cyclically. At the same time, when the object stage moves to the center of the bottom of the circular light source controller, the industrial computer will obtain the position signal of the object stage and feed back the position signal to the circular light source controller. controller, thereby triggering the circular light controller to start responding.

Preferably, when each of the light sources finds the final light source direction vector, the industrial computer controls each of the light sources to turn on individually again, and illuminates the metal component to be tested in the direction of the final light source direction vector, while each When the light source is on, the camera is triggered to collect images, and then the computer will calculate and process the collected images to determine whether there are defects in the metal workpiece to be tested; wherein, when the computer processes the images, it needs to rely on claims The method in step S2 described in 1.

Preferably, the surface of the circular light source controller facing the stage is coated with black paint, the light source is an infrared LED point light source; the camera is a 5 million high-definition 16mm fixed-focus lens, and its maximum target surface size is It is 2/3", the power of the light source is 2.1W, and the wavelength is 850nm.

Compared with the prior art, a metal workpiece defect detection method based on monocular stereo vision provided by the present invention has the following beneficial effects:

1. The metal workpiece defect detection method based on monocular stereo vision of the present invention utilizes multiple azimuths, that is, at least three light sources, and each azimuth is calibrated with the best light source direction vector, so that each metal workpiece to be tested is The azimuths are all under the final light source direction vector, and then by introducing a template workpiece without defects, by comparing the images collected under the final light source direction vector between the template workpiece and its metal workpiece to be tested, it is judged whether the metal workpiece to be tested is There are flaws, the advantages of this are:

The first point is that the light source is in at least three directions. Unlike the light under a single light source, it is difficult to cover the entire metal workpiece to be tested, and there will inevitably be too dark areas, resulting in dead angles that are not covered by the light in the subsequent collected images. Or too dark areas, so that the positions of many defects of the metal workpiece to be tested are ignored, thereby causing inaccurate detection; the invention sets light sources from multiple directions to cover the metal workpiece to be tested, so that the metal workpiece to be tested needs to be detected All the parts are irradiated by the light source, so there will be no dead angle areas when collecting images, and the detection accuracy and reliability are relatively better;

The second point is that the light source in each direction is calibrated as the best light source vector, which can be understood as the most suitable direction for each light source. The advantage of this is that there will be no overexposure or overdarkening in each direction. , so there will be no bright spots or dark spots when the subsequent image is collected, which increases the quality of the collected image, thereby further increasing the accuracy of defect detection;

The third point is that each time the metal workpiece and the template workpiece to be tested are under the direction vector of the final light source, which ensures the consistency of the background of the reference light source during comparison. Compared with the previous metal workpiece detection technology, the template workpiece used for comparison is In a fixed situation, for example: the image of the template workpiece taken during detection is preset before, instead of being collected under the same light source direction every time, so it is very different for judging the metal workpiece to be tested. Accurate, resulting in a large error rate in the final judgment result, which brings economic losses; the present invention compares the metal workpiece to be tested with the preset template workpiece under the same final light source direction vector, so that there is only one variable, and after the comparison The results are highly reliable.

2. There are many methods of light source direction calibration in the prior art. The method of the present invention to obtain the final light source direction vector in each orientation is to set a light source in each orientation, and set a calibration ball within the irradiation range of each light source, using The calibration ball is placed five times at equal intervals in the irradiation range of each light source, and the images of each calibration ball at different positions are collected, and the final light source direction vector in each orientation is obtained through an algorithm. This method can quickly obtain each orientation The final light source direction vector below, compared with other light source calibration methods in the prior art, has the advantage that: the use of equipment has no other special requirements for light source and camera, only the calibration ball is needed as a calibration tool, and the implementation process is very simple. It is necessary to use the camera to shoot the calibration ball to collect the image, and the calculation process ensures the accuracy of the calculation. The final light source direction vector obtained by using the least square method can be understood as the result of the calibration ball at different positions in each orientation. The direction of the light source calculates the result with the smallest error.

At the same time, for the light source in each direction, the calibration ball needs to be placed five times at equal intervals in the irradiation range of each light source, so as to ensure the precision and accuracy of the light source direction calibration.

3. The method for judging whether there is a defect in the metal workpiece to be tested in the present invention is: using a template workpiece to compare with the metal workpiece to be tested under the same final light source direction vector, the advantage of this is that the output result is guaranteed to be a dependent variable, and If there is a defect in the metal workpiece to be tested, it will become brighter, and the others will be uniform, thus greatly improving the accuracy of defect detection.

4. Further, after detecting whether there are defects in the metal workpiece to be tested, the mechanical arm can be added for sorting, thereby improving production efficiency, using automation to increase the quality of the workpiece, saving manpower, and reducing the defect rate of the workpiece, such as some waiting The metal workpiece to be tested is a brittle piece that has not been post-processed. Human hands may cause hand marks or pollution, and uneven manual force may cause deformation of the metal workpiece to be tested. The mechanical arm can be adjusted to a relatively balanced force to improve the quality of the workpiece, and The error rate of the robotic arm is far lower than that of humans, which increases the yield.

5. On the metal workpiece defect detection method based on monocular stereo vision, the present invention also proposes a system, which can use the assembly line of the factory or foundry and the detection system of some laboratories, and the system can be used according to the set Technical indicators require automatic detection of defects, classification of defective parts, and automatic sorting and elimination as needed.

6. In the metal workpiece defect detection method based on monocular stereo vision, the present invention also proposes a device based on the method, which uses a camera, a light source, a circular light source controller, a stage, an industrial computer and a computer, The light is emitted in three directions so that the light completely covers the metal workpiece to be tested, and the circular light source controller controls the evenly distributed light sources to turn on individually; when each light source is turned on individually, the camera collects an image each time and transmits it to the computer for analysis, thereby Find the final light source direction vector in each orientation, so the device has the function of automatically calibrating the light source direction. The circular light source controller can control the individual flashing of the light source, and the camera can collect images, so that the circular light source controller, camera , light source, and computer, the final light source direction vector of the light source in each orientation can be found, and the metal workpiece to be tested can be covered with light under the final light source direction vector of the light source on the stage.

7. Further, the metal workpiece defect detection device based on monocular stereo vision also includes a driving guide rail and a mechanical arm, and the driving guide rail can circulate the stage at the bottom of the circular light source controller and move on the side close to the mechanical arm, so as to A function of carrying the metal workpiece to be tested back and forth on the stage. The robotic arm can pick up the metal workpiece to be tested and put it on the stage, and can also remove the metal workpiece that has been tested for classification. As a result, the entire detection process is fully automated, relatively stable, and product consistency is improved; it is suitable for mass production and reduces the production cost of the enterprise.

8. The surface of the circular light source controller facing the stage is coated with black paint; the purpose is to reduce the reflection of its own light source, thereby reducing its own influence in the detection process of the metal workpiece to be tested. The light source is an infrared LED point Light source, compared with the monotonous luminous effect of traditional light sources, LED light source is a low-voltage microelectronic product that successfully integrates computer technology, network communication technology, image processing technology, embedded control technology, etc., so it is also a digital information product and is a semiconductor optoelectronic device. "High-tech and cutting-edge" technology has the characteristics of online programming, unlimited upgrades, and flexibility.

9. The camera is a 5 million high-definition 16mm fixed-focus lens, its maximum target size is 2/3", the power of the light source is 2.1W, and the wavelength is 850nm, which is the most suitable for the metal workpiece based on monocular stereo vision The requirements of the defect detection device, the higher the optical resolution, the more suitable for high-precision detection. Similarly, the size of the target surface refers to the size of the photosensitive part of the image sensor. It is generally expressed in inches, and the maximum is 2/3” The size of the target surface can obtain a relatively large amount of light and a better depth of field.

【Description of drawings】

Fig. 1 is the flow chart of the metal workpiece defect detection method based on monocular stereo vision in the first embodiment of the present invention;

Fig. 2 is the first embodiment of the present invention based on monocular stereo vision metal workpiece defect detection method, a structural schematic diagram of the method of "at least three orientations" in step S1;

Fig. 3 is a structural schematic diagram of another mode of "at least three orientations" in step S1 of the method for detecting metal workpiece defects based on monocular stereo vision in the first embodiment of the present invention;

4 is a specific flow chart of step S1 in the metal workpiece defect detection method based on monocular stereo vision in the first embodiment of the present invention;

Fig. 5 is a structural schematic diagram corresponding to the implementation of step S11 in the metal workpiece defect detection method based on monocular stereo vision in the first embodiment of the present invention;

6 is a specific flow chart of step S2 in the metal workpiece defect detection method based on monocular stereo vision in the first embodiment of the present invention;

Fig. 7 is a block diagram of a metal workpiece defect detection system based on monocular stereo vision according to the first embodiment of the present invention

Fig. 8 is a structural schematic diagram of a metal workpiece defect detection device based on monocular stereo vision in a first viewing angle according to the second embodiment of the present invention;

Fig. 9 is a structural schematic diagram of a metal workpiece defect detection device based on monocular stereo vision in a second viewing angle according to the second embodiment of the present invention;

Fig. 10 is a partial structural schematic diagram of a metal workpiece defect detection device based on monocular stereo vision according to the second embodiment of the present invention.

Explanation of the accompanying drawings:

100. system; 200. device;

10. Launch module; 20. Acquisition module; 30. Processing module; 40. Execution module; 50. Control module;

1. Camera; 2. Light source; 3. Circular light source controller; 4. Stage; 5. Drive rail; 6. Industrial computer; 7. Mechanical arm; 8. Computer.

【Detailed ways】

In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be further described in detail below in conjunction with the accompanying drawings and implementation examples. It should be understood that the specific embodiments described here are only used to explain the present invention, not to limit the present invention.

The invention provides a metal workpiece defect detection method based on monocular stereo vision, which is mainly applicable to the field of metal workpiece defect detection, assists industrial intelligent production, and provides good quality control for foundry enterprises.

Please refer to FIG. 1, the first embodiment of the present invention provides a metal workpiece defect detection method based on monocular stereo vision;

The metal workpiece defect detection method based on monocular stereo vision includes steps S1 to S3:

Step S1: take the metal workpiece to be tested as the center, set the light source in at least three directions and perform light source direction calibration to obtain the final light source direction vector in each direction;

It can be understood that the center point of the metal workpiece to be tested is used as the center point, and it depends on the size of the metal workpiece. When the field angle of the metal workpiece to be tested is relatively small relative to the light source, the metal workpiece to be tested can be regarded as a point. When the metal workpiece to be tested is large, the center point of the metal workpiece to be tested is taken as the center; the interpretation of the light source direction calibration is: to find the standard of the light source direction.

For a better understanding of step S1, please refer to Figure 2 and Figure 3, the above "at least three orientations" can be understood as three positions around the metal workpiece to be tested, which can be as shown in the figure on the metal workpiece to be tested The surrounding area, or one side thereof, is specifically set according to the situation of the metal workpiece to be tested. Exemplarily, a light source 2 is set around the metal workpiece to be tested (unlabeled), and the light source 2 is in at least three directions, wherein , when the light source 2 is set in three directions, the included angle emitted by the light source 2 is 120°; when the light source 2 is set in four directions, the included angle emitted by the light source 2 is 90°. If it is complex or requires higher precision in defect detection, more orientations are used to calibrate the direction of the light source.

The significance of performing light source direction calibration is to find the most suitable light source direction vector in each orientation, that is, the final light source direction vector, in order to obtain the optimal captured image; for example, the final light source direction vector on the left orientation is It means that the light emitted from the left is only in the direction of , the effect obtained is the best, such as no exposed surface, no shadow, etc. Those skilled in the art can easily understand that the calibration of the light source direction is one of the basic tasks for visual measurement and positioning, and whether the calibration parameters are accurate or not is directly related to the accuracy of the entire detection.

Optionally, in this embodiment, 8 azimuths under the compass coordinates are used to calibrate the direction of the light source, namely: east, west, south, north, northeast, southeast, northwest: southwest. These 8 azimuths are used to calibrate the direction of the light source, then Obtain the final light source direction vectors for 8 orientations.

And step S2: Obtain images of the metal workpiece to be tested under the final light source direction vector in at least three directions; process the image using the principle of photometric stereo vision to determine whether there is a defect in the metal workpiece to be tested;

It can be understood that the template workpiece refers to a normal and non-defective workpiece, which is used as a reference object. According to the final light source direction vector after the calibration of the light source direction in each direction in step S1, the metal workpiece to be tested and the template workpiece under the final light source direction vector The images are taken and compared, for example: the image of the metal workpiece to be tested taken under the final light source direction vectors of 8 directions acquired in this embodiment, and the template under the final light source direction vectors of 8 directions The workpieces are compared to obtain the result of whether there are defects in the metal workpiece to be tested.

Among them, "photometric stereo vision" is a classic 3D model reconstruction technology.

Wherein, the definition of "defect" can be spots, pits, scratches, chromatic aberrations, and defects on the surface of the metal workpiece to be tested.

In some embodiments, step S3 is further included: according to the result of step S2, a mechanical arm is used to classify and grasp the metal workpieces to be tested.

Please refer to Figure 4, where step S1: take the metal workpiece to be tested as the center, set the light source in at least three directions and calibrate the direction of the light source to obtain the final light source direction vector in each direction, specifically including the following steps:

Step S11: Set up a light source in each of at least three directions; thus obtain light sources in at least three directions, introduce calibration balls and place them five times at equal intervals within the light source irradiation range of each direction, and for all directions The calibration sphere under the light source is collected, thereby obtaining at least 15 pairs of calibration sphere images;

Step S12: Identify the center point and the highlight point of the calibration ball in each calibration ball image, and calculate the spatial position relationship between the center point and the highlight point; regard the calibration ball as an orthogonal projection model and combine the principle of photometric stereo vision to calculate get the light source direction vector in each azimuth; and

Step S13: Using the light source direction vector obtained from each calibration ball image, use the least square method to obtain the final light source direction vector in each orientation.

Please refer to Fig. 5, wherein for step S11 proposed: "A light source is set in each orientation in at least three orientations; thereby obtain the light source of at least three orientations, introduce five calibration balls in each of the light sources of at least three orientations Place the light sources in three directions at equal intervals, so as to collect at least 15 calibration ball images in at least three light source directions", where the Phong model is the first influential light illumination model proposed in real graphics. Only the reflection of objects to direct light is considered, and the ambient light is considered constant. The reflected light between objects is not considered, and the reflected light between objects is only represented by ambient light.

It can be understood that the calibration ball (unlabeled) used is optionally a silicon nitride ceramic ball, and its surface material is high-gloss black. Since there are 8 directions in this embodiment, a light source is provided in each of the 8 directions. 2. The method of using eight light source azimuth calibration is: first place a calibration ball at any position within the light source irradiation range corresponding to light source 2 in one azimuth, collect an image of the calibration ball under the illumination of light source 2, and then equidistantly Replace the positions four times, so that a total of 5 calibration ball images under the illumination of light source 2 are collected, and then 5 calibration balls are placed within the irradiation range of the light source corresponding to another orientation, and then 5 images of calibration balls illuminated by light source 2 are collected, namely silicon nitride The image of the ceramic ball; since the present embodiment can optionally be 8 light sources 2, it is necessary to perform 8 rounds in sequence, and place the calibration ball in 5 different positions in 8 orientations; 5 photos are collected in each round, that is, Collect at least 40 calibration ball images under 8 light source direction vectors;

Wherein for step S12 described "using the Phong model (Phong: Phong model) based on the orthogonal projection model in combination with the principle of photometric stereo vision, thereby calculating the light source direction vector", this step adopts the Phong model to replace the traditional photometric stereo vision Lambertian model. Because the Lambertian body model of traditional photometric stereo vision assumes that under the Lambertian body reflection model, the brightness value of a point on the surface of an object is only related to its surface reflectance and surface normal vector, and this relationship is linear. However, in reality, the surface of most objects is neither a pure specular surface nor a pure diffuse reflective surface, but a mixed reflective surface containing these two components. Therefore, the present invention replaces the Lambertian body model with the Phong model, considers the linear combination of diffuse reflection, specular reflection and ambient light, reflects the material characteristics of the object more truly, and is closer to reality, specifically expressed as:

The reflection map equation described by the Phong model under the orthogonal projection is:

Among them, n represents the normal vector of the surface of the object. n _h represents the specular reflection direction vector of the object surface. Where k is the specular index factor, and K∈Z ⁺ , Z+ represents the set of all positive integers, and the value of k can be obtained from images with known surface shapes of similar materials, where E is the objective function of the Phong model.

Among them, for step S13, the specific method of "obtaining the final light source direction vector in each orientation by using the least squares method to obtain the light source direction vector obtained from each calibration ball image" is as follows:

When the calibration ball is placed at position 1 in the camera image field of view, 8 images of light source 2 illuminating the calibration ball are collected, and using the Phong model under the orthogonal projection, the directional unit vectors of the 8 light sources 2 are calculated, which is denoted as L ₁₁ , L ₁₂ , L ₁₃ , L ₁₄ , L ₁₅ , L ₁₆ , L ₁₇ , L ₁₈ , L ₁₁ in the label represents the light source direction unit vector of No. 1 light source 2 at position 1 (the eight light sources 2 are numbered clockwise ), and so on, the light source direction unit vectors of light source ₂ at five different positions are: L ₁₁ , L ₁₂ , L ₁₃ , L ₁₄ , L ₁₅ , L ₁₆ , L ₁₇ , _{L 18} ; L ₂₃ , L ₂₄ , L ₂₅ , L ₂₆ , L ₂₇ , L ₂₈ ; L ₃₁ , L ₃₂ , L ₃₃ , L ₃₄ , L ₃₅ , L ₃₆ , L ₃₇ , L ₃₈ ; L ₄₁ , L ₄₂ , L ₄₃ ,L ₄₄ ,L ₄₅ ,L ₄₆ ,L ₄₇ ,L ₄₈ ; L ₅₁ ,L ₅₂ ,L ₅₃ ,L ₅₄ ,L ₅₅ ,L ₅₆ ,L ₅₇ ,L ₅₈ , for these five light source direction unit vectors, suppose The optimal direction vector is L, and the method of using the least squares method to solve the unit vector of the final light source direction is as an example:

For the No. 1 infrared LED point light source, the final light source direction unit vector is solved as:

Simplifies it to:

AL=H

understandable: among them,

Finally, using the least squares method to solve, we can get:

L＝(A ^T A) ^-1 A ^T H

According to this method, the final direction unit vectors of the infrared LED point light sources in 8 directions can be calculated according to the least square method. It can be understood that the final direction unit vectors are .

Please refer to Fig. 6, wherein in step S2, "preset the template workpiece, obtain the image of the metal workpiece to be tested and the template workpiece under the final light source direction vector in at least three directions; use the principle of photometric stereo vision to digitize the image And process comparative analysis to judge whether there is a defect in the metal workpiece to be tested. The specific steps are:

Step S21: acquiring at least 3 images of the metal workpiece to be tested and the preset template workpiece under at least three final light source direction vectors;

Step S22: According to the image obtained in step S21, use the Phong model based on the orthogonal projection model to obtain the surface normal vectors of the metal workpiece to be tested and the template workpiece, and establish a gradient map respectively, and perform Gaussian transformation on the gradient map to obtain a curvature map. and performing morphological processing on the curvature map; and

Step S23: Bolb analysis is performed on the morphologically processed curvature maps of the metal workpiece to be tested and the preset template workpiece to output whether there is a defect in the metal workpiece to be tested.

Wherein, the "template workpiece" is a metal workpiece without any defects on the surface, which is consistent with the type and size of the metal workpiece to be tested.

Among them, "Blob analysis" is to analyze the connected domain of the same pixel in the image, and the connected domain is called Blob; the color spots in the image after binarization processing can be considered as Blob; the Blob analysis tool can be separated from the background target, and can calculate the number, position, shape, direction and size of the target, and can also provide the topological structure among the related spots. In the process of processing, instead of analyzing individual pixels one by one, the row of the image is operated; Each row is encoded with run-length to represent the adjacent object range; this algorithm greatly improves the processing speed compared with the pixel-based algorithm.

It can be understood that step S21 is the image collected under the final light source direction vector of each orientation obtained after step S13 is completed. Since there are a total of 8 orientations in this embodiment, the images of the metal workpiece to be measured are eight There are eight images of the measured template workpiece, so sixteen images are obtained.

The "gradient map" described in step S22 is the gradient magnitude feature map of the image. When there are edges in the gray value of the collected image, there must be a larger gradient value. On the contrary, when there are relatively smooth parts in the image, When the gray value changes small, the corresponding gradient is also small. In image processing, the modulus of the gradient is referred to as the gradient, and the image composed of the image gradient becomes a gradient image. When obtaining the gradient image, a small area template can be used for convolution. To calculate, exemplary such as: Sobel operator (Sobel: a term of edge detection, which can be translated as Sobel), Robinson operator (Robinson: Robinson), Laplace operator (Laplace: Laplace Si), etc., using Gaussian transformation to obtain the curvature diagram means: the geometric meaning of the curvature map after Gaussian transformation is the limit of the local area of the curved surface, so in this embodiment, the curvature map after Gaussian transformation reflects the local curvature of the gradient image , that is, the data reflecting the change of the gradient image. Finally, the curvature map is subjected to morphological processing. Those skilled in the art can easily understand that the meaning of the morphological processing is to observe and process the image, so as to achieve the purpose of improving image quality.

Finally, Blob analysis will be performed on the morphologically processed curvature maps of the metal workpiece to be tested and the template workpiece, and the image of the template workpiece can be used as a background to compare the image of the metal workpiece to be tested, so as to compare whether there is a defect in the metal workpiece to be tested.

Please refer to FIG. 7 , in order to better solve the above problems, the present invention proposes another solution as follows, a metal workpiece defect detection system 100 based on monocular stereo vision, which is used for defect inspection of the metal workpiece to be tested, including:

Emitting module 10: used for emitting light to the metal workpiece to be tested from at least three directions;

Acquisition module 20: for collecting the image of the metal workpiece to be tested under the irradiation of the emission module 10;

Processing module 30: used to calculate the final light source direction vector of the metal workpiece to be tested in each orientation of the emission module 10 in each orientation, and calculate and judge whether there is a defect in the metal workpiece to be tested;

Executing module 40: classifying the metal workpiece to be tested;

The control module 50 : according to the result of the processing module 30 , is used to control the transmitting module 10 and the execution module 40 to perform corresponding actions.

It can be understood that the metal workpiece defect detection system 100 based on monocular stereo vision provided by the present invention, the metal workpiece defect detection system 100 based on monocular stereo vision utilizes the metal workpiece defect detection system 100 based on monocular stereo vision provided by the present invention The overall steps of the detection method or the combination of single steps, the system can be applied to various terminal devices such as servers, computers, cameras, mobile phones, tablet computers, etc., and is realized by hardware and software; to detect defects on metal workpieces.

In order to better solve the above problems, the present invention proposes another solution as follows. The present invention also provides a storage medium, the storage medium or the processor stores a computer program, and when the storage medium program is running, the storage medium is controlled. Or the computer program to be executed by the processor needs to execute the above-mentioned metal workpiece defect detection method based on monocular stereo vision.

The second embodiment of the present invention provides a metal workpiece defect detection device based on monocular stereo vision. The illustrated device is only a schematic structural diagram set up for the convenience of explaining the method described in the first embodiment of the present invention, and is not intended to limit the invention The structure, connection relationship and positional relationship of the environment and even the shape of the environment.

Please continue to refer to FIG. 8 and FIG. 9, the metal workpiece defect detection device 200 based on monocular stereo vision provided by the second embodiment includes a camera 1, a light source 2, a circular light source controller 3, a stage 4, a driving guide rail 5, Industrial computer 6, mechanical arm 7, computer 8;

Among them, the industrial computer 6 is electrically connected to the driving guide rail 5 and the mechanical arm 7, and the industrial computer 6 can control the operation of the driving guide rail 5, so that the stage 4 can move circularly on the driving guide rail 5, and the industrial computer 6 controls the mechanical arm 7 to grab and wait. The metal workpiece to be tested in the inspection area is placed on the stage 4, and under the action of the driving guide rail 5, the stage 4 is translated to the area that can be photographed by the camera 1, that is, the position where the stage 4 stops is just in front of the camera. 1, at the same time, when the stage 4 reaches the set position, the industrial computer 6 will obtain the position signal of the stage 4 and feed back the position signal to the circular light source controller 3, and then the circular light source controller 3. Control the evenly distributed light sources 2 to be powered on so as to turn on individually; the light source 2 can emit light from at least three directions so that the light completely covers the metal workpiece to be tested, and the camera 1 collects an image each time when each light source 2 is turned on independently Transmission to the computer 8 for analysis, thereby finding the best light source direction in each orientation, the vector represents the direction in the coordinate system, so the best light source direction mentioned here is the final light source direction vector in the first implementation, for convenience Understand, the following are all described with the best light source direction.

Further, the industrial computer 6, the circular light source controller 3, the camera 1, and the light source 2 are all electrically connected in sequence, wherein the camera 1 and the light source 2 are both located on the side of the circular light source controller 3 facing the stage 4, and the camera 1 is located at the center of the surface of the circular light source controller 3, and the light source 2 is set in a circular matrix centered on the camera 1. The height of the circular light source controller 3 from the stage 4 is not limited, only the following functions need to be realized That is: when the metal workpiece to be tested is placed on the stage 4, the light source 2 can cover the metal workpiece to be tested with light in different directions, and the camera 1 can collect the illuminated image of the metal workpiece to be tested in real time.

Further, the number of light sources 2 can optionally be set to 8. When the above-mentioned circular light source controller 3 receives the position signal, the 8 light sources 2 will be turned on and off in turn, and they will be sent to the objects to be tested on the stage 4 from different directions. Metal workpieces are illuminated;

It can be understood that: sequentially on and off in this embodiment shows that multiple light sources 2 can be arranged in a sequence. The first light source 2 is turned on individually, and the rest of the light sources 2 are turned off. When the second light source 2 is turned on individually, the rest of the light sources 2 are turned off...the cycle continues in sequence until the last one in the sequence is turned on, that is, it is turned on and off sequentially.

In some special embodiments, after the above-mentioned single light source 2 is turned on and off sequentially, two light sources 2 are turned on and off sequentially, or three light sources 2 are turned on and off sequentially.

It can be understood that when the circular light source controller 3 controls the uniformly distributed light sources 2 to turn on and off in turn, and when the computer 8 analyzes the images collected by the camera 1 each time each light source 2 is turned on individually, it is the light source for the light source 2. The process of direction calibration needs to rely on the method of step S1 in the first embodiment.

In the second implementation, the method corresponding to the first step S1 is implemented, specifically: the light source 2 in each orientation is preset with a calibration ball set in the manner described in step S1, and the industrial computer 6 controls each light source 2 to be turned on separately. , the camera 1 will collect the calibration sphere image, and the computer 8 will analyze the calibration sphere image collected each time to find the best light source direction, and the analysis will be repeated in turn, so that the light source 2 in the 8 directions can find the best light source direction at this time, The analysis of the control cycle of the industrial computer 6 ends.

Furthermore, when the eight light sources 2 all find the best light source direction, the industrial computer 6 continues to control the eight light sources 2 to turn on and off individually in turn. At this time, the metal workpiece to be tested is irradiated with the best light source vector in eight directions, and at the same time Trigger the camera 1 to collect images, and then the computer 8 will calculate and process the collected images to determine whether there is a defect in the metal workpiece to be tested. When the computer 8 processes the images, it needs to rely on the steps described in the first embodiment Method in S2:

In the second implementation, the method of step S2 is correspondingly implemented, specifically: after each light source 2 is in the best light source direction, the camera 1 continues to collect 8 images of the best light source directions, which are not calibration ball images, and the calibration The ball image is used to find the best light source direction under each light source 2, and the image is the image of the metal workpiece to be tested, and is used to detect whether the metal workpiece to be tested has defects. According to the method described in the first implementation, the collected Using the Phong model under the orthogonal projection model to calculate the surface normal vector of the metal workpiece to be detected, and then use the surface normal vector to calculate the surface gradient information; according to the surface gradient information, the surface gradient information is convolved with the Gaussian derivative to obtain The Gaussian curvature of the surface, and then convert the Gaussian curvature of each point on the image into a gray value, that is, convert the Gaussian curvature of the surface of the object to be measured into a curvature image. Perform edge detection on the curvature image, then remove the interference points, fill and expand the closed area, then segment the threshold, and analyze the Blob to realize the defect detection of the metal workpiece to be tested. It should be noted that this technology is easy for those skilled in the art. Known technical means, so the first embodiment and the second embodiment of the present invention do not emphasize the part of the above-mentioned step S2.

The processed result is further fed back to the industrial computer 6, that is, the information on whether the metal workpiece to be tested has defects is fed back to the industrial computer 6, and the industrial computer 6 controls the driving guide rail 5 to move the loading platform 4 to the side close to the mechanical arm 7, At the same time, the industrial computer 6 will control the mechanical arm 7 to grab the metal workpiece to be tested, so as to perform classification processing. For example, two areas are set, one is the classification area for the defective metal workpiece to be tested, and the other is the If there is no defect in the classification area of the metal workpiece to be tested, if the result processed by the computer 8 is that the metal workpiece to be tested has a defect, the industrial computer 6 will control the mechanical arm 7 to grab it and place it in the defective classification area, and vice versa However; after the end, the industrial computer 6 controls all the above-mentioned operating elements to reset.

The necessary structure and operation mode of the metal workpiece defect detection device 200 to be tested based on monocular stereo vision have been described above, and the models and features of the necessary structure are further described below:

Please refer to Fig. 10, the surface of the circular light source controller 3 facing the stage 4 is coated with black paint (not shown), the purpose is to reduce its own light source reflection, thereby reducing its own influence on the detection process of the metal workpiece to be measured ; Reduce the background interference of the area photographed by the camera 1, and achieve the purpose of simply segmenting the metal workpiece to be tested.

Optionally, the light source 2 is an infrared LED point light source. The advantage of the infrared LED point light source is that it reduces the interference of ambient light; the image acquisition device can be used in a natural environment, not limited to a dark room environment.

Optionally, the circular light source controller 3 controls the light source 2 in an automatic mode, and the circular light source controller 3 controls the eight light sources 2 arranged in a circular matrix to turn on and off.

Optionally, camera 1 is optionally a 5 million high-definition 16mm fixed-focus lens, and its maximum target size is 2/3”; the specific model can be MER-131-210U3M(-L)NIR, and the sensor is 1/2 "COMS, optical resolution is 1280*1024, pixel size is 4.8*4.8um;

It can be understood that in a machine vision system, the lens is equivalent to the human eye, and its main function is to focus the target image on the photosensitive area array of the sensor. Therefore, the higher the optical resolution, the more suitable for high-precision detection. Similarly, the target surface The size refers to the size of the photosensitive part of the image sensor. Generally expressed in inches, and 2/3" can obtain a relatively large amount of light and a better depth of field.

Optionally, the model of the light source 2 is OSRAM; SFH4232A, the power is 2.1W, and the wavelength is 850nm.

Those skilled in the art should also understand that if the method or device described in the present invention is simply changed, combined with functions added to the above-mentioned method, or replaced on the device, such as the replacement of the model material of each component, the use of Replacement of the environment, simple replacement of the positional relationship of each component, etc.; or the product it constitutes is integrated; or detachable design; where the combined components can form a method/equipment/device with specific functions, with such a method/ Equipment/devices that replace the methods and devices of the present invention also fall within the protection scope of the present invention.

Compared with the prior art, a metal workpiece defect detection method based on monocular stereo vision provided by the present invention has the following beneficial effects:

1. The metal workpiece defect detection method based on monocular stereo vision of the present invention utilizes multiple azimuths, that is, at least three light sources, and each azimuth is calibrated with the best light source direction vector, so that each metal workpiece to be tested is The azimuths are all under the final light source direction vector, and then by introducing a template workpiece without defects, by comparing the images collected under the final light source direction vector between the template workpiece and its metal workpiece to be tested, it is judged whether the metal workpiece to be tested is There are flaws, the advantages of this are:

The first point is that the light source is in at least three directions. Unlike the light under a single light source, it is difficult to cover the entire metal workpiece to be tested, and there will inevitably be too dark areas, resulting in dead angles that are not covered by the light in the subsequent collected images. Or too dark areas, so that the positions of many defects of the metal workpiece to be tested are ignored, thereby causing inaccurate detection; the invention sets light sources from multiple directions to cover the metal workpiece to be tested, so that the metal workpiece to be tested needs to be detected All the parts are irradiated by the light source, so there will be no dead angle areas when collecting images, and the detection accuracy and reliability are relatively better;

The second point is that the light source in each direction is calibrated as the best light source vector, which can be understood as the most suitable direction for each light source. The advantage of this is that there will be no overexposure or overdarkening in each direction. , so there will be no bright spots or dark spots when the subsequent image is collected, which increases the quality of the collected image, thereby further increasing the accuracy of defect detection;

The third point is that each time the metal workpiece and the template workpiece to be tested are under the direction vector of the final light source, which ensures the consistency of the background of the reference light source during comparison. Compared with the previous metal workpiece detection technology, the template workpiece used for comparison is In a fixed situation, for example: the image of the template workpiece taken during detection is preset before, instead of being collected under the same light source direction every time, so it is very different for judging the metal workpiece to be tested. Accurate, resulting in a large error rate in the final judgment result, which brings economic losses; the present invention compares the metal workpiece to be tested with the preset template workpiece under the same final light source direction vector, so that there is only one variable, and after the comparison The results are highly reliable.

2. There are many methods of light source direction calibration in the prior art. The method of the present invention to obtain the final light source direction vector in each orientation is to set a light source in each orientation, and set a calibration ball within the irradiation range of each light source, using The calibration ball is placed five times at equal intervals in the irradiation range of each light source, and the images of each calibration ball at different positions are collected, and the final light source direction vector in each orientation is obtained through an algorithm. This method can quickly obtain each orientation The final light source direction vector below, compared with other light source calibration methods in the prior art, has the advantage that: the use of equipment has no other special requirements for light source and camera, only the calibration ball is needed as a calibration tool, and the implementation process is very simple. It is necessary to use the camera to shoot the calibration ball to collect the image, and the calculation process ensures the accuracy of the calculation. The final light source direction vector obtained by using the least square method can be understood as the result of the calibration ball in different positions in each orientation. The direction of the light source calculates the result with the smallest error.

At the same time, for the light source in each direction, the calibration ball needs to be placed five times at equal intervals in the irradiation range of each light source, so as to ensure the precision and accuracy of the light source direction calibration.

3. The method for judging whether there is a defect in the metal workpiece to be tested in the present invention is: using a template workpiece to compare with the metal workpiece to be tested under the same final light source direction vector, the advantage of this is that the output result is guaranteed to be a dependent variable, and If there is a defect in the metal workpiece to be tested, it will become brighter, and the others will be uniform, thus greatly improving the accuracy of defect detection.

4. Further, after detecting whether there are defects in the metal workpiece to be tested, the mechanical arm can be added for sorting, thereby improving production efficiency, using automation to increase the quality of the workpiece, saving manpower, and reducing the defect rate of the workpiece, such as some waiting The metal workpiece to be tested is a brittle piece that has not been post-processed. Human hands may cause hand marks or pollution, and uneven manual force may cause deformation of the metal workpiece to be tested. The mechanical arm can be adjusted to a relatively balanced force to improve the quality of the workpiece, and The error rate of the robotic arm is far lower than that of humans, which increases the yield.

5. On the metal workpiece defect detection method based on monocular stereo vision, the present invention also proposes a system, which can use the assembly line of the factory or foundry and the detection system of some laboratories, and the system can be used according to the set Technical indicators require automatic detection of defects, classification of defective parts, and automatic sorting and elimination as needed.

6. In the metal workpiece defect detection method based on monocular stereo vision, the present invention also proposes a device based on the method, which uses a camera, a light source, a circular light source controller, a stage, an industrial computer and a computer, The light is emitted in three directions so that the light completely covers the metal workpiece to be tested, and the circular light source controller controls the evenly distributed light sources to turn on individually; when each light source is turned on individually, the camera collects an image each time and transmits it to the computer for analysis, thereby Find the final light source direction vector in each orientation, so the device has the function of automatically calibrating the light source direction. The circular light source controller can control the individual flashing of the light source, and the camera can collect images, so that the circular light source controller, camera , light source, and computer, the final light source direction vector of the light source in each orientation can be found, and the metal workpiece to be tested can be covered with light under the final light source direction vector of the light source on the stage.

7. Further, the metal workpiece defect detection device based on monocular stereo vision also includes a driving guide rail and a mechanical arm, and the driving guide rail can circulate the stage at the bottom of the circular light source controller and move on the side close to the mechanical arm, so as to A function of carrying the metal workpiece to be tested back and forth on the stage. The robotic arm can pick up the metal workpiece to be tested and put it on the stage, and can also remove the metal workpiece that has been tested for classification. As a result, the entire detection process is fully automated, relatively stable, and product consistency is improved; it is suitable for mass production and reduces the production cost of the enterprise.

8. The surface of the circular light source controller facing the stage is coated with black paint; the purpose is to reduce the reflection of its own light source, thereby reducing its own influence in the detection process of the metal workpiece to be tested. The light source is an infrared LED point Light source, compared with the monotonous luminous effect of traditional light sources, LED light source is a low-voltage microelectronic product that successfully integrates computer technology, network communication technology, image processing technology, embedded control technology, etc., so it is also a digital information product and is a semiconductor optoelectronic device. "High-tech and cutting-edge" technology has the characteristics of online programming, unlimited upgrades, and flexibility.

9. The camera is a 5 million high-definition 16mm fixed-focus lens, its maximum target size is 2/3", the power of the light source is 2.1W, and the wavelength is 850nm, which is the most suitable for the metal workpiece based on monocular stereo vision The requirements of the defect detection device, the higher the optical resolution, the more suitable for high-precision detection. Similarly, the size of the target surface refers to the size of the photosensitive part of the image sensor. It is generally expressed in inches, and the maximum is 2/3” The size of the target surface can obtain a relatively large amount of light and a better depth of field.

The above are only preferred embodiments of the present invention, and are not intended to limit the present invention. Any modification, equivalent replacement and improvement made within the principle of the present invention shall be included in the protection scope of the present invention.

Claims (10)

1. A metal workpiece defect detection method based on monocular stereoscopic vision is used for carrying out defect inspection on a metal workpiece to be detected, and is characterized by comprising the following steps:

step S1: setting light sources in at least three directions by taking a metal workpiece to be measured as a center, and calibrating the directions of the light sources to obtain a final light source direction vector in each direction; and

step S2: presetting a template workpiece, and acquiring images of a metal workpiece to be detected and the template workpiece in at least three directions under final light source direction vectors; the image is digitized and processed and contrasted and analyzed by adopting the principle of photometric stereo vision so as to judge whether the metal workpiece to be detected has defects or not.

2. The method for detecting defects of a metal workpiece based on monocular stereovision as claimed in claim 1, wherein in step S1, the light source direction calibration is performed in at least three directions with the metal workpiece to be detected as the center, so as to obtain the final light source direction vector in each direction, which comprises the following steps:

step S11: providing a light source in each of at least three orientations; thereby obtaining light sources in at least three directions, introducing calibration balls and placing the calibration balls five times in the light source irradiation range of each direction at equal intervals, and collecting the calibration balls under the light sources in all directions, thereby obtaining at least 15 calibration ball images;

step S12: identifying a central point and a highlight point of a calibration ball in each calibration ball image, and calculating the spatial position relationship of the central point and the highlight point; the calibration ball is taken as an orthogonal projection model and combined with the photometric stereo vision principle, so that a light source direction vector under each direction is obtained through calculation; and

step S13: and obtaining a final light source direction vector under each direction by using a least square method according to the light source direction vector obtained by each calibration sphere image.

3. The method for detecting defects of a metal workpiece based on monocular stereovision as set forth in claim 1, wherein in step S2, a template workpiece is preset, and images of the metal workpiece to be detected and the template workpiece under at least three azimuthal final light source direction vectors are obtained; the method adopts the principle of photometric stereo vision to carry out numeralization, processing and contrastive analysis on the image so as to judge whether the metal workpiece to be detected has defects, and specifically comprises the following steps:

step S21, acquiring at least 3 images of the metal workpiece to be detected and the preset template workpiece under at least three final light source direction vectors;

step S22: according to the image obtained in the step S21, utilizing a Phong model based on an orthogonal projection model to obtain surface normal vectors of the metal workpiece to be detected and the template workpiece, respectively establishing a gradient map, carrying out Gaussian conversion on the gradient map to obtain a curvature map, and carrying out morphological processing on the curvature map; and

step S23: performing Bolb analysis on the curvature maps of the metal workpiece to be detected and the preset template workpiece after morphological processing to output a result of whether the metal workpiece to be detected has defects.

4. The method for detecting defects of a metal workpiece based on monocular stereovision as claimed in claim 1, further comprising a step S3 based on the step S2, specifically comprising the steps of:

step S3: and according to the result of the step S2, classifying and grabbing the metal workpiece to be detected by using a mechanical arm.

5. The utility model provides a metal work piece defect detecting system based on monocular stereoscopic vision for carry out defect inspection to the metal work piece that awaits measuring, its characterized in that includes:

a transmitting module: the device is used for emitting light rays to the metal workpiece to be detected from at least three directions;

an acquisition module: the device is used for acquiring an image of the metal workpiece to be detected under the irradiation of the transmitting module;

a processing module: the device is used for calculating the final light source direction vector of each orientation of the transmitting module to the metal workpiece to be detected in each orientation, and calculating and judging whether the metal workpiece to be detected has defects or not;

an execution module: classifying the metal workpiece to be detected;

a control module: and according to the result of the processing module, the transmitting module and the executing module are controlled to perform corresponding actions.

6. A storage medium or a processor storing a computer program, wherein when the storage medium program is executed, the computer program controlling the storage medium or the processor to execute needs to execute a method for detecting defects of a metal workpiece based on monocular stereoscopic vision as set forth in any one of claims 1 to 4.

7. A metal workpiece defect detection device based on monocular stereoscopic vision is used for carrying out defect inspection on a metal workpiece to be detected and is characterized by comprising a camera, a light source, a circular light source controller, an objective table, an industrial personal computer and a computer; the industrial personal computer is electrically connected with the camera, the light source, the circular light source controller, the object stage and the computer, the light source and the camera are both positioned on one surface of the circular light source controller facing the object stage, the camera is positioned in the center of the circular light source controller, and a metal element to be detected can be placed on the object stage;

when a metal workpiece to be detected is placed on the objective table, the industrial personal computer controls the circular light source controller to start responding, and the circular light source controller can control each light source to be sequentially and independently lightened; when each light source is independently lightened, the camera collects one image each time and transmits the image to the computer for analysis, and therefore a final light source direction vector under each direction is found;

wherein, when the computer analyzes the image, the method of step S1 in claim 1 is used.

8. The apparatus according to claim 7, further comprising a driving rail and a robot arm, wherein the stage is fixedly connected to the driving rail, and the industrial personal computer is capable of controlling the driving rail to operate, so that the stage can move circularly on the driving rail, and meanwhile, when the stage moves to the center of the bottom of the circular light source controller, the industrial personal computer obtains a position signal of the stage and feeds the position signal back to the circular light source controller, so as to trigger the circular light source controller to start responding.

9. The metal workpiece defect detection device based on monocular stereoscopic vision of claim 7, wherein when each light source finds a final light source direction vector, the industrial personal computer controls each light source to be independently turned on again, the metal element to be detected is irradiated in the direction of the final light source direction vector, and simultaneously when each light source is turned on, the camera is triggered to perform image acquisition, and then the computer performs calculation processing on the acquired image to judge whether the metal workpiece to be detected has defects;

wherein, when the computer processes the image, the computer needs to rely on the method in step S2 of claim 1.

10. The apparatus for detecting defects of metal workpieces based on monocular stereoscopic vision as claimed in claim 7, wherein the surface of the circular light source controller facing the objective table is coated with black paint, and the light source is an infrared LED point light source; the camera is a 500-ten-thousand high-definition 16mm prime lens, the maximum target surface size is 2/3 ″, the power of the light source is 2.1W, and the wavelength is 850 nm.