Machine vision systems are high-tech systems integrating modern computer, optics, and electronic technologies. Machine vision technology uses computers to process images captured by the system, analyze the information within, make corresponding judgments, and then issue control commands to the equipment. The specific application requirements of machine vision systems vary widely, and the vision systems themselves may take many different forms, but they all include the following processes:
Image acquisition uses a light source to illuminate the object or environment being observed, and then acquires images through an optical imaging system. The optical images are then converted into digital images using a camera and image acquisition card. This is the front end and information source of a machine vision system.
Image processing and analysis: Computers use image processing software to process and analyze images, extracting useful information. Examples include identifying open circuits in PCB board images, defects in textile images, and text within document images. This is the core of the entire machine vision system.
The information obtained from image processing is ultimately used to judge the object (the object being measured, the environment) and generate corresponding control commands, which are then sent to the appropriate mechanisms. For example, in the captured image of a part, the dimensions of the part are calculated to see if they conform to the standard; if they do not, an alarm is issued, and the part is marked or removed.
Throughout the process, the information of the object being measured is reflected as image information, which is then analyzed to obtain feature description information. Finally, judgments and actions are made based on the obtained features. A typical machine vision system generally includes: a light source, an optical imaging system, a camera, an image acquisition card, an image processing hardware platform, image and visual information processing software, and a communication module.
In general, a successful machine vision system needs to focus on solving key technologies in several aspects, including image acquisition (including light source, optical imaging, digital image acquisition and transmission), and image processing and analysis.
Lighting Design
Illumination is a crucial yet often overlooked aspect of machine vision systems. Its design is a critical step in the overall system design, directly impacting the system's success and performance. Because illumination directly affects the system's raw input, it has a direct impact on the quality of the input data. The light source is not merely for illuminating objects; effective lighting design can highlight the features to be detected while suppressing unwanted interfering features, greatly facilitating subsequent image processing. Conversely, an inappropriate lighting scheme can cause uneven image brightness, increased interference, and difficulty in distinguishing effective features from the background, making image processing extremely difficult, or even impossible.
Lighting design mainly includes three aspects: the light reflection and transmission characteristics of the light source, the target, and the environment, and the structure of the light source. Because the objects being measured, the environment, and the detection requirements vary greatly, there is no universal machine vision lighting equipment. Lighting solutions need to be designed for each specific case, taking into account the optical characteristics of the object and features, distance, background, and specifically selecting the light intensity, color and spectral composition, uniformity, shape of the light source, and illumination method according to the detection requirements.
Lighting design is a highly complex task, requiring not only theoretical knowledge and analytical skills but also frequent experimentation and adjustments. The saying, "The light source is the benchmark, lighting is the art," highlights the crucial role of lighting design in machine vision systems. This has spurred the emergence of manufacturers renowned for their light sources, such as CCS, Moritex, and Dongguan Technology. Domestic system integrators like Lingyun have also begun developing their own independent light source products.
Optical Imaging Systems and Cameras
In machine vision systems, the lens is analogous to the human eye, its primary function being to focus the optical image of the target onto the photosensitive array of the image sensor (camera). All image information processed by the vision system is obtained through the lens, and the quality of the lens directly affects the overall performance of the vision system. Once information is severely lost in the imaging system, it is extremely difficult to recover it in subsequent stages. Therefore, the appropriate selection of the lens and the design of the imaging optical path are among the key technologies of a vision system.
Lens imaging inevitably involves some degree of distortion. Significant distortion can greatly trouble the vision system, requiring careful consideration during imaging design. This includes selecting lenses with minimal distortion and ensuring the effective field of view is limited to the center field of view with minimal distortion. Another characteristic of lenses is their spectral characteristics, primarily influenced by the interference properties of the lens coating and the absorption properties of the materials. It is essential to ensure that the highest resolution light emitted by the lens matches the illumination wavelength and the wavelength received by the CCD device, while maximizing the lens's transmittance for that wavelength. Using appropriate filters in the imaging system can achieve certain special effects. Furthermore, the design of the imaging optical path must also take into account the influence of various stray lights.
A camera is a photoelectric conversion device that transforms the optical image formed by an optical imaging system into video/digital electrical signals. A camera typically consists of a core photoelectric conversion device, peripheral circuitry, and output/control interfaces. Currently, the most commonly used photoelectric conversion device is the CCD, which uses electrical charge as a signal, unlike other devices that output current or voltage signals. In the 1990s, a new type of image sensor emerged: CMOS (Complementary Metal-Oxide Semiconductor) cameras.
In addition to examining the photoelectric conversion device, factors such as system speed, detection field of view, and the required accuracy of the system should also be considered when evaluating a camera.
The analog video signal output by the camera cannot be directly recognized by the computer. An image acquisition card digitizes the analog video signal through quantization processing, creating a digital image that the computer can directly process, and provides a high-speed interface with the computer. Image acquisition cards need to acquire large amounts of image data in real time and must work in coordination with the camera to complete specific tasks. In addition to A/D conversion, image acquisition cards also have other functions, including:
● Receives high-speed data streams from digital cameras and transmits them to system memory via a high-speed computer bus;
● Multi-channel image reception, processing, and reconstruction;
● Perform functional control on the camera and other system modules.
Image and visual information processing
The front-end components of the aforementioned machine vision system, including light sources, lenses, and cameras, prepare materials for the image and visual information processing module. This module is the key and core of the machine vision system; it detects specific targets and features through image processing, analysis, and recognition. This module comprises two parts: machine vision processing software and a processing hardware platform. The visual processing software can be divided into two levels: image preprocessing and feature analysis and understanding. Image preprocessing includes image enhancement, data encoding, smoothing, sharpening, segmentation, denoising, and restoration processes to improve image quality. Image feature analysis and understanding involves detecting the target image and calculating various physical quantities to obtain an objective description of the target image, mainly including image segmentation and feature extraction (geometric shape, boundary description, texture characteristics), etc.
Commonly used algorithms in machine vision include: search, edge detection, blob analysis, caliper tool, optical character recognition, and color analysis.
Currently, competition in machine vision software has shifted from pursuing functionality to competing on the accuracy and efficiency of algorithms. There are already vendors specializing in providing vision software or development kits. While conventional machine vision software development kits can all provide the aforementioned functions, their detection performance and computational efficiency vary significantly. Excellent machine vision software can quickly and accurately detect target features in images and is highly adaptable to different image types; while poor software suffers from slow speed, inaccurate results, and poor robustness.
From a hardware platform perspective, improvements in computer CPUs and memory have provided excellent support for vision systems; multi-core CPUs combined with multi-threaded software can significantly increase speed. With the development of DSP and FPGA technologies, embedded processing modules are gaining increasing attention due to their powerful data processing capabilities, integration, modularity, and the fact that they do not require complex operating system support.
In general, machine vision is a highly integrated opto-mechatronic-computer system, and its performance is not determined by any single component. Even if every component is perfect, it doesn't necessarily guarantee satisfactory final performance. System analysis and design are the most challenging and fundamental aspects of machine vision system development, and areas where many developers lack expertise, requiring urgent improvement.
In addition, in field applications, vibration, dust, and electromagnetic interference can seriously affect the operation of the system, and these are issues that should be considered during design and development.
Currently, embedded systems, represented by smart cameras, are favored by many experts due to their many unique advantages. The distributed network composed of highly modular and inexpensive vision sensors presents us with an exciting picture.
However, the most worrying aspect of the machine vision industry chain is that some fundamental technologies and components, such as image sensor chips for cameras and advanced lenses, still rely entirely on foreign products. Domestic machine vision manufacturers are still basically at the application level of development, which is very detrimental to the popularization and promotion of this technology in my country.
