What Is Machine Vision?

Machine vision is a branch of artificial intelligence that is developing rapidly. Simply put, machine vision is the use of machines instead of human eyes for measurement and judgment. Machine vision systems use machine vision products (that is, image pickup devices, divided into two types: CMOS and CCD) to convert the captured object into an image signal and send it to a dedicated image processing system to obtain the morphological information of the captured object. Information such as brightness and color is converted into digital signals; the image system performs various operations on these signals to extract the characteristics of the target, and then controls the field equipment operations based on the discrimination results. ^[1]

Today, China is becoming one of the most active areas in the development of machine vision in the world, and its applications cover various industries of the national economy such as industry, agriculture, medicine, military, aerospace, meteorology, astronomy, public security, transportation, security, and scientific research. The important reason is that China has become the processing center of the global manufacturing industry.

Machine vision

A typical industrial machine vision system includes:

The machine vision detection system uses a CCD camera to convert the detected object into an image signal and send it to a dedicated image processing system. Based on the pixel distribution, brightness, color and other information, it converts it into a digital signal. The image processing system performs various operations on these signals. To extract the characteristics of the target, such as area, number, position, and length, and then output the results according to the preset tolerance and other conditions, including size, angle, number, pass / fail, presence / absence, etc., to realize automatic identification function .

A typical machine vision system includes the following five blocks:

In machine vision systems, getting a high-quality processable image is crucial. The success of the system must first ensure good image quality and obvious features. The failure of a machine vision project is mostly caused by poor image quality and inconspicuous features. To ensure a good image, you must choose a suitable light source.

Basic elements of light source selection:

Contrast: Contrast is very important for machine vision. The most important task of lighting for machine vision applications is to maximize the contrast between features that need to be observed and image features that need to be ignored, thereby easily distinguishing features. Contrast is defined as a sufficient amount of gray difference between a feature and its surrounding area. Good lighting should ensure that features that need to be detected stand out from other backgrounds.

Brightness: When choosing two light sources, the best choice is to choose the brighter one. When the light source is not bright enough, three bad situations may occur. First, the camera's signal-to-noise ratio is not sufficient; due to the insufficient brightness of the light source, the contrast of the image is necessarily insufficient, and the possibility of noise in the image increases immediately. Secondly, the brightness of the light source is not enough, and the aperture must be increased to reduce the depth of field. In addition, when the brightness of the light source is insufficient, random light such as natural light will have the greatest impact on the system.

Robustness: Another way to test a light source is to see if the light source is least sensitive to the position of the component. When the light source is placed in different areas or different angles of the camera's field of view, the resulting image should not change accordingly. A highly directional light source increases the possibility of specular reflection on the highlight area, which is not conducive to the subsequent feature extraction.

A good light source needs to be able to make the features you are looking for very obvious. In addition to the camera being able to capture the part, a good light source should be able to produce maximum contrast, sufficient brightness, and insensitivity to changes in the position of the part. With the light source selected, the rest of the job is much easier. The specific light source selection method also lies in the practical experience of the experiment.

In the production process of cloth, highly repeatable and intelligent tasks such as cloth quality inspection can only be completed by manual inspection. Many modern inspection lines can often see inspection workers to perform this process, giving enterprises While increasing huge labor and management costs, it still cannot guarantee a 100% inspection pass rate (that is, "zero defects"). Testing the quality of cloth is repetitive labor, error-prone and inefficient.

In foreign countries, the popularity of machine vision is mainly reflected in

The research of machine vision started from the study of the world of building blocks made up of polyhedra by American scholar LR Roberts in the mid-1960s. The techniques of preprocessing, edge detection, contour composition, object modeling, and matching used at that time have been applied in machine vision since then. Roberts uses a bottom-up approach in image analysis. The edge detection technology is used to determine the contour line, and the area analysis technology is used to divide the image into regions composed of pixels with similar gray levels. These technologies are collectively called image segmentation. The purpose is to describe the analyzed image with contour lines and regions in order to compare and match the models stored in the machine. Practice has shown that it is too difficult to use only bottom-up analysis. It is necessary to use top-down analysis, that is, to divide the target into several sub-targets, and use heuristic knowledge to predict the object. This is consistent with the bottom-up and top-down methods used in speech understanding. In the study of image understanding, A. Guzman proposed the use of heuristic knowledge to show that the method of interpreting contour drawings using symbolic processes does not need to resort to numerical calculation programs such as least square matching.

In the 1970s, machine vision formed several important research branches: target-oriented image processing; parallel algorithms for image processing and analysis; extracting three-dimensional information from two-dimensional images; sequence image analysis and motion parameter evaluation; visual knowledge Representation; knowledge base of visual system, etc.

Achilles' heel of machine vision: According to MIT Technology Review, researchers from Google and the OpenAI Institute have discovered a weakness in machine vision algorithms: machine vision can be disturbed by modified images, and Humans can easily find modifications to these images. ^[7]

The main applications of machine vision are detection and

Machine vision based dashboard integrated intelligent test system

EQ140- automobile instrument panel assembly is an instrument product produced by an automobile company in China. The instrument panel is equipped with speed odometer, water temperature meter, gasoline meter, ammeter, signal alarm lamp, etc. The production volume is large, and it needs to be done once before leaving the factory. Final quality inspection. The testing items include: detecting the indication errors of five instrument pointers such as speedometers; detecting whether 24 signal warning lights and several lighting 9 lights are damaged or missing. Generally, manual visual inspection is adopted to check, which has large errors and poor reliability, and cannot meet the needs of automated production. The intelligent integrated test system based on machine vision has changed this situation, realized intelligent, fully automatic, high-precision, and fast quality inspection of the instrument panel assembly, overcomes various errors caused by manual inspection, and greatly improves inspection. effectiveness.

The entire system is divided into four parts: an integrated multi-channel standard signal source that provides an analog signal source for the instrument panel, a dual-coordinate CNC system with image information feedback positioning, a camera image acquisition system, and a master-slave parallel processing system.

Automatic sheet surface damage control system

The surface quality of metal plates, such as large power transformer coils, flat wire radios, hazy leather, etc., has high requirements, but the original detection methods that used manual visual inspection or a dial indicator with a dial gauge were not only subject to subjective factors, but also New scratches may be drawn on the tested surface. The automatic detection system for the surface of metal plates uses machine vision technology to automatically inspect metal surface defects and detect them at high speed and accuracy during the production process. At the same time, the use of non-angular measurement prevents the possibility of new scratches. The working principle is shown in Figure 8-6. In this system, a laser is used as the light source, and the stray light around the laser beam is filtered by a pinhole filter. The beam expander and collimator make the laser beam parallel. And uniformly illuminate the surface of the inspected metal plate with an incident angle of 45 degrees. The metal plate is placed on the inspection table. The inspection table can be moved in three directions of X, Y, and Z. The camera uses a TCD142D 2048 line Chen CCD, and the lens uses an ordinary camera lens. The CCD interface circuit uses a single-chip microcomputer system. The host PC mainly completes image preprocessing and classification of defects or depth calculation of scratches, etc., and can display the detected defects or scratch images on the display. The CCD interface circuit and the PC communicate bi-directionally through the RS-232 port, combined with the asynchronous A / D conversion method, which constitutes human-computer interactive data collection and processing.

This system mainly uses the self-scanning characteristics of the linear array CCD and the X-direction movement of the steel plate to be inspected to obtain three-dimensional image information of the surface of the metal plate.

Automotive body inspection system

100% on-line inspection of the dimensional accuracy of the 800 series automobile body contours by the British ROVER Automobile Company is a typical example of machine vision system used in industrial inspection. And a CCD camera to detect 288 measurement points on the body shell. The car body is placed under the measuring frame and the precise position of the car body is calibrated by software.

The calibration of the measuring unit will affect the accuracy of the detection and is therefore particularly valued. Each laser / camera unit is calibrated offline. At the same time, there is a calibration device that is calibrated with a coordinate measuring machine in the offline state, which can perform online calibration of the camera.

The detection system detects three types of bodies at a speed of one body every 40 seconds. The system compares the test results with the qualified dimensions extracted from the CAD model and the measurement accuracy is ± 0.1mm. ROVER's quality inspectors use this system to judge the dimensional consistency of key parts, such as the overall appearance of the body, doors, glass windows, etc. Practice has proved that the system is successful and will be used for the body inspection of other cars in ROVER's system.

Banknote printing quality detection system:

The system uses image processing technology to compare and analyze more than 20 characteristics (number, braille, color, pattern, etc.) of banknotes on the banknote production line to detect the quality of banknotes, replacing the traditional method of human eye recognition.

Intelligent traffic management system:

By placing a camera on the main traffic route, when there is an illegal vehicle (such as running a red light), the camera takes the license plate of the vehicle and transmits it to the central management system. The system uses image processing technology to analyze the captured pictures and extract the license plate number It is stored in the database and can be retrieved by managers.

metallographic analysis:

Metallographic image analysis system can accurately and objectively analyze the matrix structure, impurity content, and tissue composition of metals or other materials, and provide a reliable basis for product quality.

Medical image analysis:

Automatic classification and counting of blood cells, chromosome analysis, cancer cell identification, etc.

^[8]

Machine vision has the following development trends. ^[9]

Because machine vision systems can quickly obtain a large amount of information, and are easy to process automatically, it is also easy to integrate with design information and process control information. Therefore, in the modern automated production process, people use machine vision systems for working condition monitoring and finished product inspection. And quality control.

But machine vision technology is more complicated, and the biggest difficulty is that the human vision mechanism is unclear. One can use introspection to describe the process of solving a problem, and then use computer simulation. But although every normal person is a "visual expert", it is impossible to use introspection to describe his visual process. So building a machine vision system is a very difficult task.

It can be expected that with the maturity and development of machine vision technology itself, it will be more and more widely used in modern and future manufacturing enterprises.