CN104966300B - Bearing roller image detecting system and method and image detection device - Google Patents

Abstract

The invention discloses a bearing roller image detection method, which includes the following steps: firstly, according to different shooting environments, corresponding image parameters are debugged; secondly, according to the related image parameters obtained in the previous step, the standard parts are extracted The feature points determine the standard of image recognition; finally, extract the feature points of the tested part and compare them with the feature points of the standard part to make a judgment.

Description

Bearing roller image detection system and method and image detection device

technical field

The invention relates to an image detection system, in particular to a bearing roller image detection system and method.

Background technique

Image recognition activities are often performed in daily life. For example, we can distinguish between tables and chairs, we can distinguish our parents from a very young age, and we can read normally. These are all abilities we take for granted. Before the advent of computers, no one expressed surprise. Nor has anyone noticed that human pattern recognition is a worthy subject of study. After the emergence of computers, artificial intelligence began to develop, and pattern recognition became a hot topic. When scientists discovered that it was so troublesome to realize human pattern recognition ability with machines, people realized the difficulty of this problem.

30 years ago, the digital computer appeared. Its powerful numerical computing ability made people expect that it would have the same non-numerical computing ability as the human brain. However, people’s early optimistic expectations did not become a reality. Compared with In the case of cash computers, the mechanism of the human brain is more complex. The fact that a computer surpasses humans in terms of numerical computing ability does not mean that its general computing ability is also very strong. How to realize non-numerical computing ability on a computer is just a matter of artificial intelligence and important task of machine pattern recognition.

In the late 1960s, digital image processing had a relatively complete system and formed a new discipline. In the 1970s, digital image processing technology developed rapidly, the theory and method were further improved, and the scope of application was wider. During this period, image processing was mainly related to the research of pattern recognition and image understanding systems, such as text recognition, medical image processing, and remote sensing image processing. From the late 1970s to the present, various application fields put forward higher and higher requirements for digital image processing, which promoted the development of this subject to a more advanced direction. Especially in terms of scene understanding and computer vision (ie machine vision), image processing has been developed from two-dimensional processing to three-dimensional understanding or interpretation. In recent years, with the rapid development of computers and other related fields, such as the development of image representation, scientific computing visualization, multimedia computing technology, etc., digital image processing has changed from a specialized research field to a scientific research and man-machine A ubiquitous tool in the interface.

With the rapid development of microelectronics integration technology, microcomputer technology, and digital technology, the research and development of data recording and storage instruments have also developed faster. Major companies have invested heavily in research and development of such instruments one after another. Occupy this technical field.

The data recording system began in the 1950s. In 1956, the United States first studied the test system used in the military ^[3] to meet the data collection and test tasks that many traditional methods cannot complete. Around the late 1960s, a complete set of data acquisition products entered the market abroad. Most of the data acquisition equipment and systems at this stage are dedicated systems. In the mid-to-late 1970s, with the development of microcomputers, a data acquisition system integrating collectors, instruments and computers was born, which was gradually divided into two categories during the development process, one is laboratory data acquisition systems, and the other is laboratory data acquisition systems. The other is the industrial field data acquisition system. As far as the bus used is concerned, the laboratory data acquisition system mostly uses a parallel bus, and the industrial field data acquisition system mostly uses a serial data bus. Since the 1990s, due to the continuous improvement of integrated circuit manufacturing technology, a high-performance, high-reliability single-chip data acquisition system has emerged. The combination of computer, single-chip microcomputer and large-scale integrated circuit is managed by software, which reduces the cost and volume of the system, multiplies the functions, and greatly enhances the data processing capacity. At this stage, the data acquisition system adopts a more advanced module structure, using ARM, DSP, FPGA and other high-performance chips, according to different application requirements, by simply adding and changing modules, combined with system programming, the system can be expanded and modified. Quickly form a new system. Throughout the domestic and foreign data logger market, the current data logger mainly has the following characteristics and development trends:

1. Intelligent, with microprocessor as the core, so as to realize high-performance, multi-loop detection system. Some recorders also have networking and communication functions, historical data query functions, and can print various curves and data in real time or on time;

2. The signal input is universal, the user can arbitrarily choose voltage, current, thermal resistance, thermocouple, etc. as the input signal type, which also makes the application of the instrument more extensive;

3. Large storage capacity and diverse methods. Most data loggers use large-capacity RAM, SRAM, FLASH, floppy disk or even hard disk to save data. The capacity of these storage methods is far greater than that of traditional recorders;

4. The display schemes are diversified, most of which are liquid crystal display and digital tube display, among which liquid crystal display has TFT color and black and white liquid crystal.

Contents of the invention

The technical problem to be solved by the invention is to provide a simple image detection method for bearing rollers.

In order to solve the above technical problems, the present invention provides a bearing roller image detection method; including the following steps:

Firstly, adjust the corresponding image parameters according to different shooting environments; secondly, extract the feature points of the standard part according to the relevant image parameters obtained in the previous step, and determine the standard of image recognition; finally, extract the feature points of the tested part , compared with the feature points of the standard parts to make a judgment.

As an improvement to the bearing roller image detection method described in the present invention: the image parameters are threshold binarization data, image expansion coefficient data, image corrosion coefficient data, and Canny edge extraction data in an image acquired in a certain environment The relevant value of .

As a further improvement to the bearing roller image detection method described in the present invention: during the processing of the threshold value binarization data, when a single high-frequency noise in the image includes a range of one to four pixels, it is set as the binary value of this processing. The value range of the value threshold; during the image expansion coefficient data processing, the image after the threshold binarization data processing is expanded, and when the small black particle interference in the image disappears completely, its expansion number is used as the current value. Expansion coefficient: When the image corrosion coefficient is processed, the image after the image expansion coefficient data processing is corroded, and when the number of pixels contained in a single scar in the image is greater than or equal to four, the number of times of image corrosion at this time is used as Image corrosion factor.

As a further improvement to the bearing roller image detection method described in the present invention: the feature point extraction step of the standard part is to perform rough positioning, fine positioning, preprocessing, and calculation of the area of the inner and outer rings of the image of the standard part in sequence; The point extraction step is to perform rough positioning, fine positioning, preprocessing, inner and outer circle breakpoint judgment, inner and outer circle trend judgment, inner and outer ring area and inner ring area calculation and comparison for the recorded image of the tested part in sequence.

As a further improvement to the bearing roller image detection method described in the present invention: in the image collected by the rough positioning, the area where the bearing roller is located is clipped; Scan points, and obtain the leftmost, rightmost, uppermost, and lowermost ends of the target bearing roller profile; the preprocessing is to perform threshold binarization and image expansion on the images determined by precise positioning according to the determined image parameters. Coefficient, image corrosion coefficient, Canny edge extraction; the inner and outer circle breakpoint judgment, the inner and outer circle trend judgment, the inner and outer circle area and the inner circle area calculation are all carried out by scanning the outer circle contour and the inner circle contour of the image.

As a further improvement to the bearing roller image detection method described in the present invention: the feature points of the measured piece are compared with the feature points of the standard piece to judge according to the order of inner and outer circle breakpoint judgment, inner and outer circle trend judgment, and inner and outer ring area comparison. .

As a further improvement to the bearing roller image detection method described in the present invention: during the steps of determining the break point of the inner and outer circles, the trend of the inner and outer circles, the area of the inner and outer rings, and the calculation steps of the area of the inner ring, scan the contour of the outer circle and the inner circle of the image After the point is selected, the following judgments are made respectively: Judgment of inner and outer circle breakpoints: If the next scanning point cannot be found in the point scanning step, it is judged that there is a breakpoint, and it is confirmed that the tested part is unqualified, and the next step is not carried out. One-step judgment; if there is no breakpoint, judge the trend of inner and outer circles; judge the trend of inner and outer circles: if there is an inward bending phenomenon in the point scanning step, it is judged that the trend of the inner and outer circles does not conform to the circle, and it is confirmed that it is unqualified, and no longer Carry out the next step of judgment; if the trend conforms to a circle, compare the area of the inner and outer circles; compare the area data of the standard part with the area of the inner and outer circles of the tested part and the area of the inner circle respectively, if the area comparison data is not within a certain range If the data is within the error range, it is judged to be unqualified, and if the data is within a certain error range, it is judged to be qualified.

As a further improvement to the bearing roller image detection method described in the present invention: calculation of the area between the inner and outer circles: in the process of scanning the outer contour, from the current scanning point to the direction of the inner circle along the ordinate to find the area passing through the black The edge point behind the area, the difference between this point and the ordinate of the scanning point on the outer contour; the scanning point of the outer contour continues to scan points, and then calculate the difference in the same way; accumulate these differences, that is, the inner and outer circles The area of the inner circle; the calculation of the area of the inner circle: in the process of scanning the inner contour, the ordinate of the accumulated scanning point is the area of the inner circle.

As a further improvement to the bearing roller image detection method described in the present invention: in the step of scanning points, the whole bearing roller is divided into 4 parts, and each part is Circle; the starting point of the scanning point is two symmetrical pixels; one of the pixels is used as the upper left and lower left The starting point of the circle, look for the edge point from the upper left, and look for the edge point from the lower left; use another pixel corresponding to it as the upper right and lower right The starting point of the circle.

The bearing roller image detection device includes an image acquisition device and an image analysis device: the image acquisition device includes a dark room, the lower side of the dark room is a bearing roller support platform; the bearing roller support platform in the dark room is directly above , set the camera; the side wall of the dark room is a white curtain wall with a radian; the lighting device is set directly above the dark room, and the lighting device includes an LED light head, and the LED light head includes a main shooting light in the middle and a surrounding There are at least four auxiliary shooting lights evenly distributed among the main shooting lights; a soft light cloth is arranged directly below the main shooting lights and the auxiliary shooting lights.

The present invention can be applied in detecting the same type of bearing rollers, and the camera with the USB interface is used to collect images, and the collected images are captured into the computer through OpenCV. In image preprocessing, threshold value binarization, morphological transformation dilation, morphological transformation erosion and Canny edge extraction are used to simplify the information of the image itself and highlight the characteristic value of the image. Use image feature point extraction and comparison to perform image recognition and make judgments.

The bearing roller image detection method has the characteristics of simple mechanical structure, easy operation, fast identification speed, wide identification range, large number of single identifications, and the like. The whole system has low cost, high efficiency, strong versatility and anti-interference, and has broad market prospects.

Description of drawings

The specific implementation manners of the present invention will be described in further detail below in conjunction with the accompanying drawings.

Fig. 1 is a schematic structural diagram of an image acquisition device of the present invention;

Fig. 2 is a schematic structural view of the main shooting light 2 and the auxiliary shooting light 3 in Fig. 1;

Fig. 3 is the white curtain wall 10 in Fig. 1 .

detailed description

The invention provides a bearing roller image detection system and method and an image detection device.

The image detection method of a bearing roller includes the following steps:

First, adjust the corresponding image parameters for different shooting environments;

Secondly, according to the relevant image parameters obtained in the previous step, the feature points of the standard part are extracted to determine the standard of image recognition;

Finally, extract the feature points of the tested part, and compare them with the feature points of the standard part to make a judgment.

Its image parameters are threshold binarization data, image expansion coefficient data, image corrosion coefficient data, and relevant values of Canny edge extraction data in the image acquired in a certain environment. When the threshold value binarization data is processed, when a single high-frequency noise in the image includes a range of one to four pixels, it is set as the value range of the binarization threshold value for this processing. When processing the image expansion coefficient data, the image after the threshold value binarization data processing is subjected to expansion processing, and when the interference of small black particles in the image completely disappears, the expansion times are used as the expansion coefficient at this time. When processing the image corrosion coefficient, the image after the image expansion coefficient data processing is subjected to corrosion processing. When the number of pixels contained in a single scar in the image is greater than or equal to four, the number of image corrosions at this time is used as the image corrosion coefficient. Canny edge extraction data is an existing technology.

The feature points of the tested part are compared with the feature points of the standard part to judge according to the order of inner and outer circle breakpoint judgment, inner and outer circle trend judgment, and inner and outer ring area comparison. The steps of judging breakpoints of inner and outer circles, judging the trend of inner and outer circles, calculating the area of inner and outer circles and the area of inner circles are carried out when scanning points of the outer circle contour and inner circle contour of the image, and record all the data of the scanning points during one sweep , after the data is analyzed, the judgments are made in the following order:

Judgment of inner and outer circle breakpoints: If the next scanning point cannot be found in the point scanning step, it is judged that a breakpoint has occurred, and the tested part is confirmed to be unqualified, and the next step is not judged; if there is no breakpoint, Then judge the trend of inner and outer circles;

Judgment of the trend of inner and outer circles: If there is an inward bending phenomenon in the scanning point step, it is judged that the trend of the inner and outer circles does not conform to the circle, and it is confirmed that it is unqualified, and the next step is not judged; if the trend is in line with the circle, the inner and outer circles are judged area comparison;

According to the area data of the standard part and the area of the inner and outer circles of the tested part and the area of the inner circle, respectively, if the area comparison data is not within a certain error range, it is judged as a substandard part; if the data is within a certain error range, It is judged to be qualified.

Recording all the data when scanning points can reduce the amount of data recording and reduce the error of subsequent data processing. The above-mentioned error range is limited according to the machining accuracy of each bearing roller, and its value range is related to the size of the bearing roller. The larger the cross-section of the bearing roller, the larger the error range, generally 100 Between to 600 pixels; the actual value depends on the size of the bearing roller (the actual size error is reflected by the pixels, so the error range here is the number of pixels).

Embodiment 1, Figures 1 to 3 present an image detection device. If you want to obtain the best and fastest detection effect during detection, you need to set up the detected image, and select the image with the best display effect for detection, but under normal natural light conditions, it is difficult to achieve the best The effect of image capture will in turn cause image processing to be too cumbersome and affect the efficiency of image detection. Aiming at this, the present invention provides a bearing roller image detection system, which includes an image acquisition device and an image analysis device.

The image acquisition device of the present invention comprises a darkroom 1, the lower side of the darkroom 1 is provided with a bearing roller support platform, the side wall of the darkroom 1 is a white curtain wall 10 with a curvature, and the lighting is arranged directly above the darkroom 1. device, the lighting device includes an LED light head, and the LED light head includes a main shooting light 2 in the middle and four auxiliary shooting lights 3 evenly distributed around the main shooting light 2 (in order to obtain a better shooting effect, it can also be placed on the main shooting light Use more evenly distributed auxiliary shooting lights 3) around the shooting LED lights; set a layer of soft light cloth 4 directly below the five LED lamp heads (main shooting lights 2 and auxiliary shooting lights 3); place bearings in the darkroom 1 A camera 5 is arranged directly above the bearing roller pallet at the roller position. When in use, place the bearing on the bearing roller pallet, and then adjust the camera 5 accordingly so that the picture taken by the lens is the entire pallet. At this time, the camera 5 captures the image of the bearing and stores it, and the captured The image is the bearing and the bearing platform, and because the bearing platform exists, as long as the bearing is placed in the bearing platform, the images taken include the complete bearing. This setting method can directly form the rough positioning of the bearing and reduce the time of bearing inspection. Calculations.

During the image acquisition process, since the bearing rollers are generally metal objects, when they are directly photographed directly, there will be obvious reflective effects. In this case, the camera 5 will not be able to photograph the metal surface The texture will affect the detection of scars on the surface of the bearing rollers. Therefore, in the present invention, in order to obtain the best shooting effect, the bearing rollers are first built into the darkroom 1. The darkroom 1 of the present invention is front, rear, Both the left and right sides are white curtain walls 10 with radians, which achieve the effect of uniform ambient light through the reflection and scattering of light by the white background; and a camera 5 and a light source are arranged directly above the dark room 1 . Although the darkroom 1 is composed of a white curtain wall 10, which can reflect and scatter the light when shooting, but when the uneven light directly irradiates the bearing roller, shadows will still appear, and the image of the present invention During the analysis process, if there is a shadow, accurate analysis will not be possible, so it is necessary to eliminate the shadow effect; at this time, the method of multi-directional illumination is used, and the lighting device is realized by five LED lamp heads, one of which is centered, and the The other four are evenly arranged around it. When shooting at this time, it can effectively eliminate shadows and other phenomena. Its five LED lamp holders are covered with a layer of soft light cloth 4, so that the light irradiated on the surface of the bearing roller is even and soft, and the result of shooting at this time can reach the requirements of image analysis in the present invention.

After the image acquisition device performs image acquisition, the image analysis device performs image analysis and processing, and the analysis and processing steps are as follows:

After collecting the image that meets the detection requirements, it is necessary to remove the noise, strengthen the useful information, and restore the degradation phenomenon caused by the input measuring instrument or other factors. It is necessary to preprocess the image. The preprocessing steps are as follows:

Threshold binarization: An image includes target objects, background and noise. To directly extract target objects from multi-valued digital images, the most common method is to set a global threshold T, and use T to set The data of the image is divided into two parts: the pixel group larger than T and the pixel group smaller than T. Its mathematical expression is as follows:

In conjunction with the image acquisition device used in the present invention, the value range of the binarization-binarization threshold is between 100 and 180 (the determination of the specific value is the value when the small black particles just appear in the image); After comparing the images with various threshold values, it is found that when the threshold value is around 150, the binarization effect is the best. At this time, the surface contour of the bearing roller is clear, and the image of the "scar" is quite obvious, even a very small "scar"; but at this time, there will be noise effects of discontinuous black particles. For such effects, the image processed by threshold binarization is again processed by image dilation.

Image expansion processing: Morphological transformation dilation (dilation) ⊕ Use vector addition (or Minkowski set addition, such as (a,b)+(c,d)=(a+c,b+d)) to merge two sets . Expansion X⊕B is the set of sums of all vectors. The two operands of vector addition come from X and B respectively, and any possible combination is taken.

X⊕B={p∈ε ² , p=x+b, x∈X and b∈B}

The dilation operation with isotropic structuring elements can be described as a transformation that converts all background pixels adjacent to the object into object pixels.

Select the number of expansion: the purpose of morphological transformation and expansion of the image is to remove the black discontinuous small particle noise in the image. When the threshold is 130, the main interference is the black particle noise on the surface of the bearing roller. After a morphological transformation and expansion of the image, the black particle noise can be completely eliminated. After increasing the number of dilations, the characteristics of the outer contour are affected.

In the process of morphological transformation and expansion, the background white is expanded, which can eliminate the noise of small black particles to a certain extent. However, larger black particles can only be eliminated after multiple expansions.

When the number of expansions is increased in order to eliminate larger particle points, the inner circle of the roller will be significantly affected. Therefore, when adjusting the parameters, it is necessary to control the black particles to a certain size under the premise of ensuring that the "scar" is obvious; the number of expansions is to ensure that the black particles are completely eliminated without affecting the characteristics of the cross-sectional image. That is, this requires the cooperation between the binarization threshold size and the number of times of image expansion to achieve the desired effect.

Image corrosion: corrosion (erosion) Use vector subtraction for the set elements to merge two sets. Erosion is the dual operation of expansion. Neither erosion nor dilation are reversible operations.

This formula states that every point p of the image has been tested; the result of erosion consists of all points p that belong to X satisfying p+b.

Basic morphological changes can be used to find object contours in images, and it is fast. The specific implementation method is to calculate the difference between the original image and the corroded image.

Erosion is also used to simplify the structure of objects - those objects or parts of objects that are only one pixel wide will be removed. This breaks down more complex objects into simpler parts.

Select the number of times of image erosion: After the binarization and expansion of the image are carried out above, the "scar" feature that the image should have is weakened. Especially the "island" type "scar" on the surface of the roller, it is likely that after the dialogue is expanded, such "island scar" becomes smaller; this is not conducive to the subsequent edge extraction of the image.

After the binarization threshold is set to 130 and the image is expanded once, the image is corroded for different times.

The purpose of etching the white of the image is to highlight the "scar". However, it should be noted that the number of times of corrosion should not be too much. Too many times of corrosion will cause "scars" to affect the inner and outer rings of the bearing roller section itself. In addition, erosion and dilation are not reciprocal transformations - if you first erode an image and then dilate it, you will not get the original image. The resulting image will be simpler than the original, with some details removed.

Erosion first and then expansion is an important morphological transformation called opening. The opening operation of image X with respect to structural element B is denoted as XοB, defined as

Dilation followed by erosion is called closing. The closing operation of image X with respect to structural element B is denoted as X·B, defined as

According to the above description, the opening operation used in the present invention is actually an opening operation, and when using these two transformations, the expansion is performed first and then corroded. What needs to be paid attention to here is the object of expansion or corrosion. Expansion of white is equivalent to corrosion of black. Therefore, when choosing whether to open or close the operation, it can be determined through experiments.

Canny edge extraction:

Canny proposed a new edge detection method, which is optimal for step-type edges affected by white noise. The optimality of the Canny detector is related to the following three criteria:

Detection criteria: no important edges are lost, and there should be no spurious edges.

Positioning criterion: The deviation between the actual edge and the detected edge position is minimal.

Single Response Criterion: Reduce past responses to a single marginal response. This point is partially covered by the first criterion, since when there are two responses corresponding to a single edge, one of them should be considered spurious. This third criterion addresses the problem of edges affected by noise, acting against non-smooth edge detectors.

It can be found that if two black areas are very close (the distance is only one pixel), the Canny edge extraction will result in the discontinuity of the edge. This destroys the continuity of the inner and outer rings on the original bearing roller surface. Pushing back to image corrosion, it can be found that the main reason that mainly affects this phenomenon is the number of corrosions. The more times of corrosion, the larger the area of the black block and the closer the distance; so when selecting the number of image corrosions, the next Canny edge extraction should be considered to reduce the impact.

After the preprocessing, it is necessary to perform a feature point extraction step on the image. The feature point extraction step mainly includes the feature point extraction step of the standard part and the feature point extraction step of the tested part. The standard part basically uses qualified finished products for shooting. , so after storing the recorded data in the step of capturing the target part, it scans the points in the feature point extraction step, and calculates and obtains the data of the inner and outer circle area and the inner circle area after scanning the points, and stores them as standard data, while the measured Parts need to be in the feature point extraction step, after scanning points, then through the inner and outer circle breakpoint judgment, inner and outer circle trend judgment, inner and outer circle area and inner circle area error judgment steps to determine whether it is a qualified part.

The amount of data obtained from images or waveforms is considerable. For example, a text image may have thousands of data, an electrocardiogram waveform may also have thousands of data, and a satellite remote sensing image may have a larger amount of data. In order to realize classification recognition effectively, it is necessary to transform the original data to obtain the features that can best reflect the essence of classification. This is the process of feature extraction and selection. The feature extraction steps of the present invention are as follows:

Snap target part:

The process of capturing target parts is mainly to narrow the scope of detection and reduce the amount of calculated data. Because the image of the feature value to be extracted is relatively simple, the color is a single black and white; the capture target part of the present invention is mainly divided into two steps:

Coarse positioning: frame the target bearing roller in a rough square frame;

Because the position of the bearing roller under the camera 5 is roughly fixed each time, there is only a slight deviation; so in order to reduce the amount of calculation, the area of interest can be artificially selected.

Precise positioning: frame the target bearing roller in the smallest rectangular frame;

The area of the image to be processed is greatly reduced through the steps:

Step 2 is to scan points on the basis of step 1 to obtain the leftmost, rightmost, uppermost and lowermost ends of the bearing rollers. This enables the bearing rollers to be framed in a minimal matrix.

After the step of capturing the target part, after preprocessing (the data determined above are sequentially operated on the threshold value binarization data, image expansion coefficient data, image corrosion coefficient data, and Canny edge extraction data), the feature points can be determined. Finally determine whether the tested part is qualified.

Feature point determination:

The feature point is the feature that can best explain the object, and it needs to be representative, and at the same time, it must be easy to obtain. Initially set 3 kinds of characteristic points in the present invention, introduce as follows:

Feature point 1: Whether there are breakpoints in the inner and outer rings:

The image after image preprocessing (preprocessing data is determined according to the quality of the captured image, in the embodiment, the binarization threshold is 130, the number of dilation is 1, and the number of erosion is 1). If it is a standard part, there will be no discontinuity in the inner and outer rings of the bearing rollers. The occurrence of discontinuities may only occur when two black blocks are very close to each other, or when the bending angle is at an acute angle (tip), discontinuities are prone to occur. Therefore, this feature point can be used to judge whether the measured object is a standard part. That is, according to feature point 1, a 2-dimensional feature quantity can be obtained——whether there is a breakpoint in the outer circle and whether there is a breakpoint in the inner circle.

Feature point 2: Whether the trend of the inner and outer rings conforms to a circle:

This characteristic point is obvious. If it is a non-standard part (generally, the black block is glued to the inner and outer rings), when walking along the edge, there will be a more obvious inward bending phenomenon. In practical applications, it is impossible to judge 100% whether the trend is wrong, and what can be judged is only a few wrong trends. Once it can be judged that the trend is wrong, the subsequent judgment can be skipped and it can be directly concluded that the bearing roller is a substandard product, which can reduce the overall calculation amount. That is, the feature point 2 can also obtain the 2-dimensional feature quantity—whether the trend of the outer circle is correct and whether the trend of the inner circle is correct.

Feature point 3: the area between the outer ring and the inner ring and the area of the inner ring:

This feature point serves three purposes:

1. Make up for the loopholes missed by feature point 2;

If the inner and outer rings are concave inward or protruding outward, the area will have the most direct change. This can make up for the loophole that feature point 2 cannot 100% detect trend errors. The cooperation of the two can not only improve the working efficiency of the program, but also improve the accuracy.

2. Eliminate the bearing rollers whose round corners do not meet the standard (the outer ring of such bearing rollers will be smaller than the standard parts);

It is more accurate than human eyes when detecting whether the rounding of bearing rollers meets the standard. Once the various feature points of the standard parts are determined, that is, under the premise that the area between the outer ring and the inner ring is determined, the area of the bearing rollers with non-standard rounded corners will be relatively small.

3. Remove bearing rollers with black blocks (substandard products) between the outer ring and inner ring;

In addition to considering the black blocks that are glued to the inner and outer circles, isolated black blocks are also considered, which are also "scars".

When comparing this kind of interview with the distance test, it is considered that if you choose to compare the distance of each point with the standard part, the error is too small to reflect the difference. Using the area as a feature point is equivalent to accumulating the error of the distance—the error is magnified.

According to the feature point 3, a 2-dimensional feature quantity is determined.

The total set feature quantity is 6 dimensions, namely: whether there is a breakpoint in the outer circle; whether there is a breakpoint in the inner circle; whether the trend of the outer circle is correct; whether the trend of the inner circle is correct; whether the area between the outer circle and the inner circle conforms Standard; whether the area of the inner ring meets the standard.

When performing the above-mentioned scan point step, in order to achieve the best work efficiency, the present invention adopts the following scan point sequence:

Divide the entire bearing roller into 4 parts when sweeping, each part is round. The starting point of the scanning point is two symmetrical pixels. Take one of the pixels as the upper left and lower left two The starting point of the circle, look for the edge point from the upper left, and look for the edge point from the lower left; use another pixel corresponding to it as the upper right and lower right The starting point of the circle.

Block scanning has the following two advantages: 1. Guaranteed the upper left and upper right The trend of circular sweep points is mainly upward, lower left and lower right The trend of circular scanning points is mainly downward; in intelligent operation, its software programming can be simplified and the probability of errors can be reduced. 2. In case of encountering a breakpoint or trend error and jumping out of the scanning point, the work efficiency of the present invention is improved.

In order to adapt to the arc trend of the arc, the order of scanning points at different stages of the arc also needs to be changed. Because the trend of the arc is relatively flat, the difference in trend is mainly reflected in the beginning and end of the arc, so it can be roughly divided into two stages. The "watershed" of the two stages is roughly set as the point (lateral distance of the upper left semicircle/2, y). According to the point selection method, the point will be represented by "intermediate point" in the following.

For example, in the upper left of the scan When the scanning point is circular, when the scanning point just starts to scan, the main trend of the arc is upward; when the scanning point passes the middle point, the main trend of the arc is to the right. This point scanning method can avoid that when the scanning point reaches a pixel and searches for the next scanning point in a general order, it will be misjudged as a breakpoint of the arc because no qualified point can be found. Such a misjudgment Undoubtedly a big loophole. By scanning points through the above methods, such misjudgments can be avoided.

The area between the inner and outer circles described above is calculated as follows:

In the process of scanning the outer contour, from the current scanning point to the inner circle direction along the ordinate to find the edge point after passing through the black area (referring to the area without edge) (the edge point is the white point, which is the inner edge point , for unqualified parts, because there will be defects such as black blocks in the middle, when it detects the outline of black blocks, it is not a real edge point, but it is mistaken for an edge point), this point is consistent with the scan on the outer contour The vertical coordinates of the points are subtracted; the scanning points of the outer contour continue to scan points, and the difference is calculated in the same way; these differences are accumulated, which is the area between the inner and outer circles;

Calculate the area of the inner circle described above:

In the process of scanning the inner contour, the ordinate of the scanning point is accumulated, which is the area of the inner circle.

Through this area calculation, since the outer circle deformation, inner circle deformation or inner circle cannot be accurately identified, after the scanning point is completed, the area calculation result will deviate from the standard part, and it can be determined whether it is a qualified part according to this deviation .

Finally, it should also be noted that what is listed above is only a specific embodiment of the present invention. Obviously, the present invention is not limited to the above embodiments, and many variations are possible. All deformations that can be directly derived or associated by those skilled in the art from the content disclosed in the present invention should be considered as the protection scope of the present invention.

Claims (6)

1. a kind of bearing roller image detecting method；It is characterized in that：Including the steps：

First, for different shooting environmentals, corresponding image parameter is debugged out；

Secondly the associated image parameter, obtained according to previous step, the characteristic point of extraction standard part determines the mark of image recognition It is accurate；

Finally, the characteristic point of measured piece is extracted, the characteristic point of contrast standard part is judged；

Described image parameter is threshold binarization data, image expansion coefficient data, figure in the image obtained under a certain environment As etch factor data, the related value of Canny edge extracting data；

During the threshold binarization data processing, when the single high-frequency noise in image includes one to four pixel coverages, if It is set to the span of the binary-state threshold of this processing；

During described image coefficient of expansion data processing, the image after threshold binarization data processing is subjected to expansion process, works as figure When little particle of black interference is wholly absent as in, number of times is expanded as the coefficient of expansion now；

During the processing of described image etch factor, the image after the processing of image expansion coefficient data is subjected to corrosion treatment, works as image In the pixel number that is included of single scar when be more than or equal to four, regard Image erosion number of times now as Image erosion Coefficient；

The feature point extracting step of standard component for standard component image is carried out successively coarse positioning, fine positioning, pretreatment, inside and outside image Enclose areal calculation；

The feature point extracting step of measured piece for the measured piece image recorded is carried out successively coarse positioning, fine positioning, pretreatment, The judgement of inside and outside circle breakpoint, the judgement of inside and outside circle tendency, Internal and external cycle area and inner ring areal calculation are simultaneously compared.

2. bearing roller image detecting method according to claim 1, it is characterized in that：The coarse positioning is by the image of collection In, editing is carried out with the region where bearing roller；

The image that the fine positioning is determined according to coarse positioning carries out sweeping a little for contour line, and obtains the target bearing roller profile High order end, low order end, the top and bottom；

The pretreatment is to carry out threshold binarization, figure successively to the image that fine positioning is determined respectively according to the image parameter of determination As the coefficient of expansion, Image erosion coefficient, Canny edge extractings；

The inside and outside circle breakpoint judges, inside and outside circle tendency judges, Internal and external cycle area and inner ring areal calculation are by image Excircle configuration and interior circle contour sweep a progress.

3. bearing roller image detecting method according to claim 2, it is characterized in that：The characteristic point contrast standard of measured piece The characteristic point of part judges that the order judged according to inside and outside circle breakpoint, inside and outside circle tendency judges, Internal and external cycle area is compared is carried out successively.

4. bearing roller image detecting method according to claim 3, it is characterized in that：The inside and outside circle breakpoint judges, interior When cylindrical tendency judgement, Internal and external cycle area and inner ring area calculation step, swept a little for image excircle configuration and interior circle contour Afterwards, following judgement is carried out respectively：

Inside and outside circle breakpoint judges：If swept in a step, when can not find next scanning element, that is, it is determined as breakpoint occur, confirms Measured piece is unqualified part, no longer carries out next step judgement；If there is not breakpoint, inside and outside circle tendency judgement is carried out；

Inside and outside circle tendency judges：If sweeping a step interior curved phenomenon occurs, that is, it is determined as that inside and outside circle tendency does not meet circle, confirms It is unqualified part, no longer carries out next step judgement；If its tendency meets circle, Internal and external cycle area comparison is carried out；

It is not compared according to the inside and outside area of a circle of the area data of standard component and measured piece and inner circle Line Integral, if area ratio To data not in the range of certain error, then unqualified part is determined that it is, if data are in the range of certain error, judge it For pieces O.K..

5. bearing roller image detecting method according to claim 4, it is characterized in that：Area meter between the inside and outside circle Calculate：

During outline is swept, found by the current inside circular direction of scanning element along ordinate through the side after black region The ordinate of edge point, the point and the scanning element on outline carries out making poor；

The scanning element of outline continues to sweep a little, in the same way calculating difference again；

These differences are added up, as the area between inside and outside circle；

The areal calculation of inner circle：

During Internal periphery is swept, the area of the ordinate, as inner circle of cumulative scan point.

6. bearing roller image detecting method according to claim 5, it is characterized in that：It is described to sweep in a step, will be whole Bearing roller is divided into 4 parts, and every part isCircle；

The starting point swept a little is symmetrical two pixels therein；

It regard one of pixel as upper left and lower-left twoMarginal point is found in round starting point, upper left upwards, and lower-left is sought downwards Look for marginal point；It regard another corresponding pixel as upper right and bottom right twoRound starting point.