|Machine Vision Attributes & Limitations|
Video camera-based machine vision systems have been used for industrial inspection and quality control for a number of years. However, they have only recently been integrated into AIDC applications because 1D linear bar codes were scanned far more cost-effectively with laser scanners. The recent development and use of 2D matrix bar codes has rapidly driven technology refinements and cost-efficiencies in vision-based scanning used for AIDC applications.
Vision-based scanners use a CCD-based video imager, very similar to a video camera, to capture an image and convert it into a digital format. Fluorescent lighting, high-speed strobe flash, or an array of LEDs most usually provide the illumination source. Specialized electronic circuitry and/or software processes the digitized image to obtain the encoded data.
Decoding algorithms have been developed for both 1D and 2D symbologies (as well as for Optical Character Recognition applications). Because a vision-based system captures a symbol's complete image, more information is gathered than can be obtained from a single or raster-scanned laser beam. Processing this wealth of information allows for reading at lower contrast ratios and a greater ability to work around impairments.
The first machine vision scanners were sophisticated fixed-mount devices used for expensive inspection applications. However, within the last few years the widespread interest in 2D symbologies, spearheaded by the electronics and automotive industries, has resulted in rapid developments in dedicated 2D scanning systems.
Five years ago, United Parcel Service (UPS) pioneered the development of the 2D matrix code MaxiCode as well as the vision-based CCD scanner to read it, in order to encode package address and handling data that would stay with the package. UPS's fixed-mount scanner (its manufacture was later licensed to two AIDC vendors) uses an array of sensors positioned ahead of the scanner to detect the height of packages. The sensors instantly adjust the scanner's autofocus, thus achieving an unheard-of 36-inch DOF. Fixed-mount vision-based scanners (with more modest DOFs) are also used by the United States Postal Service (U.S.P.S.).
Two different handheld CCD-based image reading scanners have lately come to market that can read both 1D and 2D stacked and matrix bar codes. The scanners operate in a working zone that extends approximately three inches from target.
Vision recognition systems
Vision recognition systems are intended to capture a visual image and, through a process of feature extraction and analysis, automatically recognize application-defined marks, characters, code structures and/or other features in the image.
With developments in automation and computer aided manufacture, applications are being identified for vision recognition systems, both stand-alone and in combination with other automatic identification technologies. These are typically in process control, quality assurance, security systems, robotics and computer-aided manufacturing. For example, vision systems are being used to read surface-relief and pierced metal bar codes in various production based applications. They can also be designed to read alphanumeric characters, stamped into metal plates for example, and markings on very small components, such as semiconductor wafers.
Vision systems are generally expensive, the cost reflecting the quality, versatility and sophistication of the system optics, cameras, processing hardware and, as appropriate, system software.
Vision recognition systems may vary substantially in design, but various functions are generic to most if not all vision systems. In general a vision recognition system comprises:
A visual image is captured and digitized and features, such as edges, holes and other contours, are extracted according to a pre-defined routine either in hardware or (at some loss in speed) by software. These features are then compared with stored reference patterns or templates as a basis for achieving recognition, or analyzed to effect a classification based upon a set of parameter measurements.
Visible light is usually the means of illuminating the objects or features to be recognized, but it is possible for systems to be specified for specialist applications where the source may be beyond the visible range (infrared or ultraviolet for example).
The versatility of vision systems is a result of the flexible optical possibilities, from microscopic to telephoto, and the ability to use mirrors, lenses, filters and fibre optics to direct the light path.
The quality of the image captured by a vision recognition system is influenced by the optical system, the speed and resolution of the camera and the digitizing system. CCD solid state cameras or array sensors are typical in vision recognition systems, with array sizes typically of the order of 420 horizontal by 480 vertical pixels resolution (larger arrays are available), and electronic shutter times down to 0.5ms.
CCD line sensors, comprising a line of charge coupled sensor elements (up to 4192 pixels in length) may be used in some applications and used, with the aid of a special field assembler memory, to create a two dimensional image of a moving object or feature.
Users can expect a wave of technological improvements and cost reductions in vision-based scanning for AIDC applications over the next few years. In certain tightly tolerant reading environments, vision-based scanners are already competitive in cost/performance with laser scanning. Currently cost considerations effectively limit resolution to about 350,000 pixels (a 500-by-700 matrix), but rapid development in consumer electronics products - CCD imagers in digital photography and in high definition television - is pushing those limits. High performance 1,000-by-1,000 arrays with a resolution of one million pixels will be available in the not-too-distant future.
As microprocessors speed up so will vision-based image processing, to a point where it may even exceed the technological limits of laser.