"Camera, camera on the wall, what's the best technology of them all?"General In a machine vision system the hardware is the body and the software is the soul, which actually creates the vision system. When machine vision first emerged 20–30 years ago, there were only few vision tools available, such as edge detection, correlation The hardware was rare and expensive. The problem machine vision engineers face today is the opposite – there are hundreds of application-specific algorithms and a wide range of hardware solutions available. This article discusses the machine vision solutions used today and makes an overview of some future possibilities.. Architectures and history perspective In the early days the machine vision systems were complex parallel computing systems with technology borrowed from CCTV markets (tube cameras, room illumination lamps and rectangular pixels etc.). Today, most of the cameras in industrial use are still analogue and they are used in interlaced TV-format. Standard LVDS and RS-422 digital outputs with their infinite cable combinations are almost history in the line scan cameras, because of the Camera Link standard. The same will slowly happen to the analogue video signals with coaxial cables. The main trend in machine vision is clearly towards fully digital imaging chain. Roughly, the machine vision system architectures used today can be divided in four different categories:
The trend in the system architecture is to solve simple application-specific problems using smart sensors, while PC-based solutions remain suitable for more general needs.
PC-based machine vision solutions, which replaced DSP-based solutions some years ago, have many advantages. The performance of a common PC roughly doubles each 18 months, which gives a significant performance advance. Especially for applications that require a user interface, the PC is a very appealing choice. It provides many interesting features in the same package, such as network connections, operating systems, displays, and mature software development kits, among others. Today with fast PCI-X bus, dual processors and Linux operating system up to 8 CCIR cameras could be processed on-line with high frame rate from each camera.
Have DSP processors disappeared from machine vision world? Not at all, they are here among us – even stronger than ever. They are embedded in the camera or sensor. Among the fastest growing segments of the machine vision market are “smart cameras” and “smart sensors”. Smart cameras or compact vision sensors that combine sensor, image processing hardware and software, power supply and communications are often presented as the ultimate answer for machine vision end users. This is true within the design limits of the system. By choosing a smart camera solution, the user accepts at the same time many design conditions: only few software libraries, limited processing power, lack of the expandability and flexibility and some smart cameras have fixed lenses and integrated illumination. Because smart cameras are developed for mass markets they also have limitations related to the frame rate, resolution, I/O, communication format and special sensor requirements like sensitivity or spectrum. Should these conditions be changed, the smart camera appears to be most non-flexible and it must probably be replaced.
 Still smart cameras are right choice for many applications. Success of the smart cameras is based on easy to set up and minimal engineering time required for application development and support. Also the physical size is small. System Parts and Bottlenecks Every machine vision systems consist of parts like illumination, optics, CMOS/CCD detector, connection bus, processing unit, operating system and application software. Despite the recent development of the cameras and especially software the illumination design is still the most critical element in machine vision engineering. Reliability can be increased and programming effort decreased with proper illumination engineering. Ten years ago the available elements were spotlights, ring and back lights, mostly made from fibre optic halogen lighting or fluorescence tube. Now there are all kinds of variations on sizes, wavelengths and techniques. LED-based technology is volume illumination technology for close distances, because it has many advantages: many colours and wavelengths, narrow spectrum, powerful especially if powered by pulsing, long life-time, low heat emission and tolerate vibration Another interesting trend is the diode laser based stroboscope technology development. Currently those are powerful, very coherent and collimated illumination tools for imaging and measuring small, high-speed targets in industrial and research applications. Among digital cameras there are CMOS and CCD solutions. About 20% of the new machine vision cameras are based on CMOS detector. In the volume detector markets – like mobile phone cameras – CMOS is the clear market leader. CMOS detectors have clear benefits in high speed imaging applications. The same CMOS camera could operate in progressive scan format at user-selectable speeds. Full asynchronous image capture with programmable partial scan (region of interest) provides the flexibility to utilize the camera in variable machine vision applications. The CMOS sensor is highly sensitive in the near infrared spectrum. Best CMOS cameras utilize the true potential of CMOS functionality by employing sensors that boost the signal to noise ratio and optimize spectral response. Analogue/digital conversion and image pre-enhancement like white balancing, and more, can be built into the CMOS sensor. Still CMOS has work to do in order to beat CCD totally from the market, like "fixed pattern noise": each pixel of the CMOS sensor has its own amplifier and the amplifiers aren't all equal, and this creates a noise pattern across the image. There is also the fact that CMOS sensors are generally less sensitive than their CCD counterparts. Currently there are following digital interfaces for camera connections: categories:
Currently, there is no digital interface that is able to address all the needs of all the markets. The market share of digital interfaces is slowly increasing, and total cost of the imaging chain will soon beat the cost of the analogue. In short term future there will be also wireless camera interfaces for special industrial applications. Most probably bluethoot, but if ethernet-protocol will be down to camera interface in future, why not WLAN…? Who needs frame grabbers anymore, because there are Firewire cameras on the market? There are features what simple direct Firewire interface connection can't deliver but normal frame grabber can, like: camera interface, asynchronous, external synch, ability to buffer the image and free up the host from babysitting the video stream - features that machine vision people use that aren't supplied by 1394 solution. Future trends and applications Faster CPUs will enable more powerful vision algorithms. Also, emerging digital standards and faster interfaces will enable higher resolution and higher-speed imaging, which will result in the need for smarter software to process the increasing number of pixels. Neural network techniques are also about to become a key paradigm in machine vision that is used either to correctly segment an image in a wide variety of operational condition or to classify the detected object. Stereo and 3D-vision applications are increasing all the time. Another trend is to utilise machine vision at non-visible spectrum - measuring the light you cannot see: X-rays, UV and also IR are already utilized widely. Especially development in near-infrared and thermal imaging technology has led to totally new machine vision applications like: remote monitoring, surface inspection, night vision, medical imaging, laser beam analysis and thermal condition monitoring. Video surveillance systems are currently undergoing a transition where more and more analogue solutions are being replaced by digital. Digital technology enables audio/video data compression that minimizes transmission bandwidth and storage requirements and permits security cameras to operate on standard data networks without the expense of bulky coaxial cables. Also there is need for new machine vision algorithm area: video stream analysis while compression and multimodality image fusion utilizing multisensor information. Expertising your projects Every environment changes. Ambient changes in lighting, temperature, air quality, vibration, and hosts of other environmental factors can impact the vision system's performance. Despite these physical limitations, the smart engineer can plan for future growth or changes to a machine vision system – as long as the vision system's capabilities and limitations are fully understood. The bottom line is simple: a machine vision system is a complex system designed to solve a specific problem. It requires knowledge of cameras, optics, illumination, frame grabbing, software and processors and there aren't too many who have a thorough knowledge of these all technologies. Atostek has years of experience in this very sophisticated technology area and knowledge of key technologies and industries related to the machine vision. It is good to keep in mind that machine vision hardware generally accounts for one third to one-tenth of the cost of a total machine vision solution, with the remainder going to expert hardware and software selection, integration and programming. The cost of installing machine vision is small compared to the cost of not doing it right. The major unexpected cost can occur after the system is installed, while trying to make it work. So, the efficiency one gains by hiring a consultant is the vastly improved probability of success from day one.
Atostek provides software and system development according to the 5-step product development process. One of the core competences being image processing, image analysis, and image manipulation. Atostek's services include application analysis producing the requirement specification of the solution, specification of the proposed solution, feasibility study, architectural design, software implementation, system testing, updates and proposals for further development. Important parts of the software project may include the design of user interface, databases, transfer in IP-network, and integration to other software systems, such as production management systems. Let Atostek expertise Your project and avoid pitfalls and costly mistakes!
|
