JPS63503332A - Inspection equipment - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

[Detailed description of the invention] Inspection equipment The present invention relates to an industrial inspection device, and is particularly suitable for rapidly inspecting multiple parts. related to the device.

Industrial inspection involves the identification of products and parts moving at slow speeds on conveyors. These include separation, positioning, counting, monitoring, and measurement. Therefore, the product This has to be investigated in real time, which makes it difficult to analyze the number of pixels required. The processing rate exceeds the limit of a single typical serial processor. ing. The actual processing rate is the speed that a single central processing unit (CPU) can process. degrees, so special hardware must be designed. There wasn't.

For this purpose, the applicant has devised an image processing multiprocessor (IMF).

The IMF is supporting the Versatile Modular Eurocard VMEbus and Crate (c r a t e). Crate is a framework for image processing. A framework that holds a special set of coprocessor boards that contain system storage. copro Because the processors operate quickly, the system can perform inspections in real time. Wear. The system itself is designed to be integrated, and the transport on the vMEbus Transaction data has low overhead, so it is processed quickly.

Additionally, the memory subsystem is configured to operate at the maximum speed the vME bus is capable of. It is measured.

According to the second invention, an inspection device for continuous inspection of a plurality of products having a common feature group is provided. It is equipped with a This inspection device includes means for detecting the position of individual products and scanning means for capturing an image in the vicinity of the imaged product; and It is equipped with analytical means to analyze the importance of the characteristics identified. Further, the positioning means a selection means connected to selectively obtains data correlated to features of the region of interest; selectively operative to control the processing of data obtained from said scanning means; It is possible.

In addition, it can be used as an inspection device that continuously inspects multiple products that have a common feature group in the area of interest. The image location device used must be able to locate detection means for generating a plurality of electrical signals corresponding to the intensities of the optical signals at a plurality of positions nearby; , generates a pair of electrical signals depending on the intensity of the optical signal at a position in the vicinity of said predetermined position. A group of electrical signals weighted symmetrically according to a predetermined algorithm to a coupling means for coupling, and a differential means for generating an electrical signal according to the difference between the pair of electrical signals; An image location device is provided that includes a stage.

The IMF system uses a frame containing four image brains of 128x128 bytes. built around memory. Memory is the access speed of the VME protocol It is designed for an access speed of 150n5ec.

The purpose of the IMF system is to perform industrial inspections in real time and at a reasonable cost. That is true. Image from line scan or TV (visicon) camera is digitized and fed to a frame memory under computer control. next , these images are processed quickly by special processing boards. this special place The physical board group operates under the control of the host computer and autonomously operates via the vME bus. It has a group of coprocessors that can dynamically access image data. .

The arbitrator on the VME bus is the host processor , the image display hardware and the various coprocessors. IMF is a multiprocessor system in which many processors can perform various functions simultaneously. It is a stem. However, only one processor can access the bus at any time. deer However, this restriction is not as strict as one might think. First, provide local memory By performing pipeline processing, parallel processing of coprocessors is possible. No. Second, many real-time industrial inspection applications require speed. required, but not at speeds high enough to work effectively with this type of system. do not have. In this case, the design of the coprocessor stem is not only rigorous, but also special It required some ingenuity. Therefore, the IMF's capabilities are substantially similar to many commercially available , more expensive than cheaper systems with simple design.

IMF systems have particular application in food testing. Quite a few foods have round features ing. When inspecting, you can quickly locate the product you are inspecting and identify critical features and defects. needed to be effectively monitored. Food products are produced in large quantities on continuously moving lines. It will be done. These lines are generally 1 to 2 meters wide and are approximately the same as the width of the conveyor. There are 12 to 20 items in the direction and at a speed of 30 to 50 centimeters per second. Moving. Therefore, the rate at which the line passes through a given point is generally 20 per second. It is individual. This means that one IMF processor processes one product approximately every 50 milliseconds. This means that you need to handle The processor selects each item from the received images. must be discovered and monitored. An appropriate resolution for one item is 60 on one side. It is a square consisting of 80 pixels. Therefore, searching for items is non-trivial and monitoring is less obvious. This kind of problem requires access to many pixels, but This is not enough to require the VME bus to be paused, so the data flow cause problems.

Algorithms for product search and monitoring and processors to execute them Considerable attention has been paid to the selection of types. For inspection of round food products The first algorithm created, first executed in hardware, now Since it is generalized, it is suitable for a wide variety of products. First, round or round Products with holes can be searched quickly and easily. Second, such products are inspected. and can be closely monitored. Third, many features of non-circular products can be searched accurately. , can be monitored near or related to these features. Features of this system Symptoms are the ability to recognize certain features that can trigger other parts of the system. That's what I mean.

According to applicant's experience, holes, specs, dots, letters, corners, line intersections, etc. You can easily search for many features including: Therefore, the IMF system is extremely and can be used in a wide range of industrial search and robotics applications can. Furthermore, there may be cases where there is no need to search for the object or some features. There are also cases. This may be the case, for example, in the gripper of a robot, the bottom of a chute, etc. This is the case when the existence and location of is known. The IMF system naturally Able to deal with situations. Overall, the IMF system is rather general. Practical in utility.

The following functions are currently performed by various processors.

(1) Edge detection and orientation; (2) Radiation intensity history near special features Structure of the gram; (3) Pixels of various intensities with various tags attached Pixel counts; (4) Internally generated counts such as distances from features of the region of interest; Correlation of threshold patterns to parameters. Other features include (5) corner Structure of intensity histogram; (6) Structure of overall object intensity histogram and (7 ) construction of intensity histograms within specific regions; and (8) construction of more general gray scales. There is a scale sub-image correlation. These features allow you to quickly calculate the center of your product. , Measuring the perimeter and area For example, to measure the perimeter and area of chocolate on a biscuit. can be done. The novel features of this invention allow most processes to be done quickly and efficiently. A processor that often performs, among other things, the identification of image regions that are correlated to a given starting position. There is an auto scan. This processor is connected to an internal bus. Inside The partial bus indicates the currently accessed pixel (x, y) and coordinates (r , hold e). This is the information that correlates to the pixel (x, y) and the coordinates (r, e). This is done by providing a lookup table on the processor board that provides the information. be exposed.

This system uses a simple function that allows you to freely relocate images, so Flexible and efficient.

In addition to these advantages, the entire system is based on an FDP-11 type host processor. can be easily controlled using high-level language. There is also an interface for this Various types are made.

A 68000 or other host processor may also be used. important here The good thing is that no complicated assembler-level programming is required. Ko All that is necessary for the processor is to initialize it using the RESET signal. , starts operation with 5TART pulses, and stores certain information in specific registers and memory. It is set to relocation.

Data is read and written to other locations by the host or other processor You can also do that. Control is via minimal registers. in these registers are named after high-level languages to facilitate programming.

The coprocessor function is a special function that occupies most of the actual inspection system.

These functions are not common, so you should perform all inspection tasks yourself. I can't. Therefore, the 'glue(g1ue)1 function is required. This machine The functions are performed by the host processor. If this glue function slows down the speed , if the load is too high, additional software coprocessors can be added to the IMF system. can be added to the system. Currently, it works in conjunction with BitSlice processors. It is possible to implement a software coprocessor via EC's T-11 processor. It is planned to make this possible. Processors like this are everything for IMP. It is impossible to perform this function. This is because at least 50 machines are being produced at the same time. Unless you add a working processor to your system, it will be too slow to test. That's because there isn't. The IMF system achieves this by processing large volumes more economically. It was created to solve a type of problem. Therefore, instead of brute force solutions, intelligent It can be said that this is a solution. In addition, typically one or more software A processor is attached to the hardware.

An embodiment of the present invention will be described with reference to the accompanying drawings.

FIG. 1 is a schematic diagram of an image processing multiprocessor system.

FIG. 2 is a schematic diagram of processor II of the IMF system shown in FIG.

FIG. 3 is a schematic diagram of processor module A of processor II.

FIG. 4 is a schematic diagram of processor module B of processor II.

FIG. 5 is a schematic diagram of processor module C of processor II.

FIG. 6 is a schematic diagram of processor module D of processor III.

FIG. 7 is a schematic diagram of processor module E of processor IV.

FIG. 8 is a schematic diagram of processor module F of processor IV.

To perform an inspection task, a product must first be searched. effective testing It can be done in two steps. (1) Product search and (2) Product monitoring. The shape is For objects that can only be determined using a lookup table, a grayscale image is used. It can be done directly. In order to perform Hough transformation, It is necessary to search for the The processor ■ is designed for this purpose. It is possible to output data from special reference points within the image, including the product.

Due to the general nature of the Hough (Hou g'h) transform, the IMF checks the starting reference point. have to search. Special reference points can be searched within images containing products to A basic starting point for the inspection process can be found without additional processing. Wear.

Once the fiducial point is found in the image, we then analyze the image in its vicinity, thereby determining the It is possible to monitor objects. Processor II systematically scans the region of interest. Perform this function using the AutoScan module. image subarea There are two advantages to scanning. First, delete irrelevant areas. Increase processing speed. Since the reference point is one or other standards, many matches can be taken. In other words, comparisons can be made with previously stored data. Wear. Processor II has a module for executing a method for performing matching processing. are doing. Processor II has an internal bus that carries reference points. In particular, The partial bus carries information about the coordinates (r, e) of the current pixel relative to the reference point. child The coordinates of are obtained from a lookup table that can be loaded into RAM. at a certain time , only one reference point can be used, but a lookup table can also It is possible to have several additional information groups, and each additional information group also includes a special type of object. or are correlated with characteristics. For example, additional information can be used to describe the ideal Contains size or other detailed information. Therefore, Processor II: While looking at several objects of different types in succession, scan the image and record its output. Force information can be stored in memory. (With this configuration, when on the VME bus You can save time. ) designed as relative location i.e. RL Babas The internal (X, y)/(r, 9) information bus is used to transfer data to the seeds within Processor II. supply to each module. In particular, the neighborhood of a special reference point can be quickly compared to the value of an ideal field. Radial and angular intensity histograms can be constructed that provide a means of comparing .

For this analysis, intensity histograms of various ranges can be generated and the standard For quasi-distributions, we can create correlation values for the distribution of special intensity values. emphasis here What we want to do is reduce the amount of processing that the host processor has to do. The objective is to quickly monitor special areas of the image using a variety of standard methods. P If the image details are unknown in Processor II, the host processor should check them. Can be done.

Processor I performs the essential edge detection functions. Actually, Hou gh) is optimized to perform transformation calculations. For this purpose, we use the Hough transform calculation. It is not enough to simply calculate; the size and direction of the edge must also be determined. . Processor l is designed to determine both the size and direction of edges. . In addition, Processor I uses a new and unique method, i.e., high-level edge size used to threshold. The reason for this is that the number of edge points can be significantly reduced. Measure the speed up of the entire IMF system and increase the average edge points searched. This is to make the accuracy higher than the standard edge points. For this reason, the process In S-I, a double threshold is set for the intensity value of the pixel, and Erase points and further increase processing speed.

Processor I has a more complex autoscan unit than Processor II. Ru. In other words, for a 1x1 window configuration in the Processor II, the processor I has a 3x3 window configuration. Furthermore, the previous two pixels (i.e. , since each new pixel just requires a subwindow of IX3). Power pixel data is reduced, calculations are done in Vibration, and processed by the host processor. Furthermore, by reducing the output data to faster local memory that is also accessible You can measure the speed up.

The priority levels of the VME bus are as follows in descending order:

Host processor acting as a level 3 system controller Level 2: Hardwired, micro, including Processor I and Processor II Code coprocessor level 1: Software coprocessor containing DECT-11 processor processor Level 0: Video display circuit Software is easier to build than hardware because it is easier to build more complex functions than hardware. software coprocessors are more intelligent than hardware coprocessors. There are many cases where Therefore, hardware coprocessors are non-implantable and require It is necessary to finish the assigned operation as necessary. Therefore, the priority is set to level 2 instead of level 1. Video display circuit plyo is the lowest, so V is only used when the bus is not occupied by other processes. Images from ME bus memory can be displayed. same priority level The processors are connected in a deci-chain, and the arbitrator module The processor closest to the bus can get the bus first.

In industrial systems where algorithm creation or processing speed is not the top priority, Representation of pixels in high-level language is effective. P of a pixel in a 5x5 window The notation according to PL2 is as follows.

P15 Pi4 P13 P12 pHP16P4 P3 P2 PIO P17 P5 POPi P9 P18 P6 P7 P8 P24 P19 P20 P2I P22 P23 Location of an image (x, y) This results in a group of pixels around . The frame memory of the IMF system Mapping is done automatically with lookup tables. In the above example, 5x5 window, but the IMF uses this method to handle windows up to 7x7 in size. be able to deal with windows.

The lookup operation slows it down a bit, but it is more programmable than the above. It has a major advantage in that it is easy to process. The automatic restart adopted in this example The mapping uses absolute addresses instead of index addresses. This has the advantage of speeding up pixel access. This is for example Directly access all images in a huge memory space like 68,000, now available Not found in possible consumer systems.

The size of the lookup table required for remapping is Similarly, Y is combined with the 6-bit window replacement information. Al in the size you need. For a 128x128 frame memory, there are two backup tables are configured, each table containing 7+5-13 address bits and 7 It has 1-bit coordinate data and 1-bit overrange data. Also, 256x For 256 frame memories, the two tables have 8+6-14 address bits. and 8+1 data bits. For 128x128 frame memory, All you need is two 8Kx8 EPROMs.

The device according to the invention can be used to process signals from edge detectors. Ru. Quickly select pixels that give accurate edge orientation and position information and ignore other pixels. The processing speed can be increased by viewing. Highly accurate position and orientation information The principle in selecting pixels to feed is to look for pixels with a very uniform intensity gradient. . This can be done by thresholding a high level intensity gradient uniformity parameter or It is obtained by thresholding the low-level non-uniformity parameter. this is , taking two symmetrically weighted sums of pixels near the pixel position of interest. It is obtained by If the location has a precisely uniform intensity gradient, then The weighting can be changed so that their sums are the same. the difference between these sums results in a non-uniformity parameter, and the threshold detector set to a suitable threshold It can be detected by

This method will be explained in the case of a 3x3 neighborhood. In this case, detect edges We use the Sobel operator for . To express the pixel intensity in the vicinity, use the following Using a notation like , the (Sobel) X and y components of the intensity gradient are: be made to suffer

gx - (C + 2F + 1) - (A + 2D + G) gy - (A + 2B + C )-(C+2H+I) Moreover, the intensity gradient is expressed by the following formula.

g　　　[g　112　+g　,2　]　1/2 or an appropriate approximation formula Ru. The orientation of the edges can be determined by using the arctangent function and determining the relative values of g8 and gy. It can be inferred from

The symmetric sum of pixel values that can be used to calculate gradient uniformity is expressed as:

sl -A +C+G 10I s2 = B + D + F + H s3-4E Therefore, the non-uniformity parameter is expressed by the following equation.

The uniformity detector dictates that the edge points where noise is noticeable even if they have a good shape This is to exclude it from consideration by the object searcher. Without this, the direction of edge becomes inaccurate. Additionally, a uniformity detector speeds up the algorithm. can be done. Additionally, the uniformity detector is a Sobell edge detector. It can handle any size neighborhood, not just the 3x3 neighborhood represented by the container. Wear. Note that instead of the Sobel edge detector, for example, Prewett's 3x3 edge detector can be used. You can also use a Tsuji detector. In addition, we have a symmetric combinatorial group of weighted sums and Non-uniformity can be detected by a combination of linear and non-linear Ru. (For example, in the case of 3x3, ul+u2+u3 can be used.) For large neighborhoods in , many homogeneity operators are possible.

One function of the uniformity detector is to eliminate areas where noise is occurring. . Other features include consideration of points that are not in place, such as deformities, etc. It is to leave it out of the picture. The uniformity detector takes one more step to ensure that the edge is the center of the neighborhood Improve the accuracy of edge operations by removing cases that do not pass through the can be done. Although the offset degrades the accuracy, the uniformity operator improves the accuracy of the test. It is possible to improve the establishment of output.

The uniformity detector detects the noisy edges (or apparently the edges themselves). can be used to remove.

This noise can occur within any nearby pixel. For example, when using a Sobel detector, In this case, noise in one of the neighboring pixels (excluding the center pixel) reduces the edge orientation accuracy. let The uniformity operator attempts to detect this. (Of course, sometimes detection There are cases where it is not possible. For example, two pixels are subject to noise and the uniformity operator , the noise is canceled, but the edge orientation operator does not cancel the noise. The screener is used for pre-screening, post-screening, etc. You can do leaning or both at the same time. Parallel screening is The operation is fast, so if a dedicated VLS I chip is available, it is desirable to use it. stomach. Bliss screening and post-screening are performed by the subsequent object detector. Useful for proactively deleting processed data. In addition, for edge detection One VLSI chip and a uniformity screening chip are not provided separately. You can also go with a tip. Importantly, uniformity detectors do not take many aspects into account. This can speed up the object detection algorithm.

According to the invention, in-flight radiation histograms and correlations can be used.

Pre-selected by hardware for the computer to remove redundant information Data extraction can be performed hierarchically by defining important regions of the image This can speed up the inspection process. The computer is an image editor. It saves time because you know which features of the image to check, not just the points. Don't waste it. The computer stores the stored data and the measurements from the scanner. Recalculate the edge points based on the values. This is a background point that does not need to be considered This is because we know that.

This invention allows food products (biscuits, pizza, cakes, etc.) to be uniform in size and shape. It is particularly effective for inspecting mass-produced items such as Also, Nanbapu Can be applied to legal tasks such as rate recognition, optical mark and optical character recognition, etc. Ru. In addition to optical technology, sonar, infrared, or antennae are used to capture images. This is also possible by methods such as detection.

RL no Cji ( /Ma included 7aOt・1yLN1-Lp 8′ Enter 7” D’t! Zog IV international search report

Claims (7)

[Claims] 1. consisting of detection means for locating individual articles having common features of interest; In an inspection device that continuously inspects multiple items, scanning means for capturing an image of the region and resolving important features detected by said scanning means; and a selection means connected to the location search means, the selection means for analyzing the location of interest. processing the data obtained from said scanning means to obtain data correlated to the features; An inspection device characterized in that it is operable to selectively control. 2. The selection means includes an intensity gradient uniformity detector which acts as an intensity threshold detector. The inspection device according to claim 1, characterized in that: 3. further comprising prohibition means for prohibiting analysis of a signal obtained from the certain characteristic portion. The inspection device according to claim 2, characterized in that: 4. The prohibition means prohibits the analysis of signals with noise of a predetermined level or higher. The inspection device according to claim 3, characterized in that: 5. The prohibition means is characterized in that it prohibits analysis of signals that do not match a predetermined shape. The inspection device according to claim 3. 6. The shape is the position of the step edge relative to the center of the population of measured intensities. The inspection device according to claim 5, characterized in that the inspection device is determined by: 7. A plurality of signals corresponding to the intensities of optical signals at a plurality of positions near a predetermined position in the optical image. detection means for generating electrical signals and symmetrically weighted according to a predetermined algorithm; combining the pairs of electrical signals that have been detected at a plurality of positions in the vicinity of the predetermined position. a coupling means for outputting a pair of electrical signals depending on the intensity of the optical signal; and differential means for outputting an electrical signal depending on the difference between the electrical signals. An inspection device according to any one of claims 1 to 6.