01 April 2003

Seeing it my way

Machine vision is making more and more sense . . . now.

By Steve Geraghty


Photoelectric sensor A sensor that uses the photoelectric effect-a physical phenomenon that takes place when a material is struck by light and subsequently emits electrons. One electron emits for each photon of light absorbed.

Pixel The smallest unit on a video display screen that can be stored, displayed, or addressed. A computed picture is typically composed of an array of 450 by 300 pixels or 720 by 560 pixels or other dimensions.

Visual Basic A software program that provides a macro-type language and graphical environment. The language enables a standard graphical user interface to bolt to an existing library.

Until recently, many manufacturers were unwilling or unable to implement machine vision for quality control and inspection applications. This reluctance stemmed from the expense of machine vision solutions, the level of expertise required to use them, and a perception (sometimes real) that machine vision was unreliable.

With the advent of PC-based machine vision technology and in response to the demands for quality control at every phase of the manufacturing process, system costs have dropped significantly, and new cost-effective, user-friendly solutions have appeared on the market.

In fact, because of the wide selection of machine vision options available, the problem now is choosing which type of solution is best for your application.

Here is an overview of the range of machine vision solutions available for quality assurance and inspection applications in manufacturing, some key features, advantages and disadvantages, and guidelines for when to use each type.


Pixel sensors, sometimes called "pixel counters," or "intelligent sensors," are a step up from traditional photoelectric sensors. They typically integrate a small, monochrome camera and a processor in a small, relatively low-cost package.

Pixel sensors can perform simple, high-speed operations such as detecting the presence or absence of a part or detecting defects by comparison: template matching.

Pixel sensors are easy to set up, and some even have one-button programming to learn a part, but they cannot perform complex inspections, do not offer a great degree of flexibility, and have minimal tolerance to normal process variations.

One should consider using a pixel sensor when the inspection environment-lighting, motion, and the like-is well controlled, the inspection is simple and fast, and the inspection requires no more than a pass or fail response.

Pixel sensors are not appropriate for detailed inspections or where there is significant process variation.

Vision sensors integrate the components of a machine vision system-the sensor (camera), digitizer, processor, software, interfaces, optics (lens), and sometimes illumination-into a compact package.

Because they are programmable, these products can perform many types of inspections. Vision sensor vendors provide processing and interface software, but setting up a vision sensor can still require weeks of work.

One might use a vision sensor when the inspection task is more complex than a pixel sensor can handle, when you want more than a pass or fail from the inspection, when you need flexibility in programming, and when you can afford the required time and money to program the vision sensor.

Smart cameras are similar in construction to vision sensors. The main difference is in their programming. Instead of selecting from a comprehensive and sometimes complex set of operations, smart cameras generally require users to develop applications based on library components provided by the manufacturer plus their own algorithms.

One develops the application program on a host PC and then downloads the result to the camera. These devices require a higher level of knowledge to deploy.

One might consider a smart camera when developing one's vision application that will be deploying to multiple venues. This serves to amortize the development cost.

Example applications include image compression for surveillance cameras, reading special markings on a product, or motion detection. By amortizing the development costs, smart cameras can have a lower per-unit cost than vision sensors.

Vision software
Vision Software
Graphically select areas to be inspected, and then use a point-and-click interface to designate the operations and needed reports pertaining to that area.

Vision appliance

Vision Appliance


Vision appliances are the latest development in compact machine vision systems. While their hardware is similar to that of vision sensors or smart cameras, vision appliances perform specific inspection tasks, such as optical gauging or label inspection.

Vision appliances combine sophisticated capabilities with the ease of implementation and use of a pixel sensor, so users with no previous machine vision experience can get their applications up and running in short order.

The software of each vision appliance customizes to the unique requirements of a particular application so that setting up that application is smooth.

Because there are no programming costs, a vision appliance provides a low-cost solution to many machine vision and inspection tasks. For example, the iGauge vision appliance provides precise optical measurements of features such as point positions, lines, angles, hole diameter, or roundness.

This sort of solution costs about $3,000 and is a more logical choice than a general-purpose system for part inspections or an expensive laboratory gauging system that leverages more accuracy than needed.

Vision systems have the same elements as the integrated systems previously discussed, but they use larger and more powerful components to perform demanding vision tasks.

These systems come with various levels of integration, software, and support. Most vision systems are platformed on PC technology, often with added proprietary hardware to acquire images of the parts, such as frame grabbers, and accelerate certain vision operations.

A vision system is usually the most expensive of the solutions discussed and can require weeks or months of work to set up.

But a vision system is often the only solution that is fast and robust enough to keep pace with certain high-speed manufacturing lines and complex inspection applications.

Vision systems are highly flexible and can customize to exact user requirements. Most of the work is in programming the vision algorithms and control tasks, and some vendors have reduced this work by providing easy-to-use, rapid development software.

Software lets one graphically select areas to inspect and then use a point-and-click interface to select the operations on that area and the reported results. Software facilitates rapid prototyping and production by wrapping it with a Visual Basic or Visual C++ user interface.


Most vision systems software incorporates operations including image processing and image analysis, pattern matching algorithms, statistical process control, and control of digital and other input and output lines.

These capabilities make vision systems well suited to a range of inspection and quality assurance applications, including those found in automotive, consumer goods, textile, electronics, packaging, pharmaceutical, and plastics manufacturing.

Finding the best solution for your application should start with a careful review that asks questions such as these:

  • What measurements or inspections need to take place? Communicate this to the machine vision vendors, plant staff, and integrators so that a clear and tight specification can evolve to prevent confusion or waste.
  • What level of complexity is required for the task? Is it a simple part in place or the searching for subtle defects in parts or products that have a range of acceptable variation?
  • How fast must the inspections happen? Depending on the difficulty of the task, machine vision systems process between 5 and 1,200 parts per minute. More difficult tasks often require a full vision system.
  • What technical resources already exist on site? Is there an in-house programming staff, or will the project require a systems integrator? If resources are limited and the application is not too complex, consider a pixel sensor or vision appliance.
  • What is the format for results? Is a simple "pass" or "fail" sufficient? What kinds of communication and protocols need to connect the machine vision system to the manufacturing process?
  • How soon must the system be working? Is the system unique, or will replication of the work happen once again or many times?

Armed with a detailed assessment of your application requirements, a knowledge of the level of machine vision expertise of the people who will be using and perhaps implementing the system, and ideas about the types of systems available, specifying a system can be a rewarding experience. IT

Behind the byline

Steve Geraghty has an electrical engineering degree. He has years of experience in the process industries and in imaging. Geraghty is director of ipd, a division of Coreco. You can contact him at sgeraghty@goipd.com.

Wide-eyed growth future

The worldwide market for general-purpose machine vision (GPMV) systems for manufacturing industries will grow at a healthy compounded annual growth rate (CAGR) of 8.9% during the next five years.

The market was $860 million in 2001 and will probably be more than $1.3 billion in 2006, according to a recent ARC Advisory Group study.

OEMs that once designed their own machine vision systems now prefer GPMV-based systems. Internally developed, leading-edge machine vision systems no longer make business sense for OEMs, due to the rapid advancement in technologies on many fronts.

North America is the largest revenue generator for GPMV suppliers, followed closely by Japan. Revenues generated in Europe, the Middle East, and Africa (EMEA) are well below those in North America and Japan.

North America and EMEA will grow at a higher rate than Japan during the next five-year period because of their broad industry base. Asia will grow at a double-digit rate as investment pours in to take advantage of its lower cost structure.

Manufacturers across all industries are looking to improve quality consistency, lower production and scrap costs, manage inventory accurately, gather more data from process control, and increase product reliability.

Manufacturers have recognized that GPMV solutions can help achieve these goals, leading to significant growth for the GPMV market for many years to come.

Rapid decline in the cost of implementing GPMV solutions is providing a great incentive to users to evaluate them. Smart cameras and vision sensors are at pricing levels that make them often affordable.

Advanced networking ability in both low-end and high-end machine vision solutions will also further fuel the growth in the market.

While growth rates in the leading industries have slowed, higher growth will occur in many other industries that have begun to adopt machine vision solutions.

Significant revenues will generate from leading industries, as well as food and beverage, pharmaceutical, and cosmetic industries, in the five-year forecast period.

Source: ARC Advisory Forum