Video-Controlled Production

We see video being used as an Inescapable Data gathering technology more broadly than ever before, and across a wide range of applications. Certainly, video is used for surveillance (both in homes and across public areas), and we will later examine how video can be used by retailers in ways that now go beyond monitoring for security purposesways that allow retailers to gather customer data that was previously unavailable.

In the manufacturing sector, video is becoming a key element in process automation, rapidly optimizing many activities. What has changed recently is the move to the use of all-digital video, which in turn allows for process examination and analysis by systems as opposed to eyeballs. Use of digitized video analysis driven by manufacturing needs will in turn drive significant advances and availability of digitized video as a surveillance tool in more pervasive ways (crime, antiterror, and home security).

The coming pervasiveness of digitized video has been enabled by several factors, including the following:

• High demand for consumer charge coupled display (CCD)-based gear (primarily consumer cameras and home video camcorders) has driven up production volumes and picture quality (pixel densities) while decreasing costs of the primary component, the CCD cells.

• The integration of wireless network interfaces with digital cameras has recently become more practical.

• Advances in image-processing techniques, largely made possible by increased single-chip CPU speeds and decreased component sizes, allow a camera to do most of the digital processing "locally"i.e., within the camera itself.

Cognex Corporation is a leading supplier of video-based machine vision systems.^[9] Dr. Robert Shillman (better known as "Dr. Bob"), CEO of Cognex, describes the changes taking place in the machine vision industry:

^[9] Headquartered just outside of Boston, Massachusetts, in 1981 has shipped more than 120,000 vision systems having an aggregate value greater than $1.2 billion.

As few as four or five years ago, the largest application of machine vision was in semiconductor manufacturing, where the manufacturing speeds and the physical tolerances were beyond those of human vision and attention span. Today, however, machine vision is used in a wide variety of "everyday" factory applications ranging from ensuring that labels and caps are properly placed on medicine bottles, to ensuring that all of the seat mounting bolts are present in the floor pans of trucks. The drivers for machine vision on the factory floor are the need to reduce errors, to reduce scrap, to reduce liability and, and in general, to reduce manufacturing costs, all while increasing overall product quality and increasing production speeds.

The advances in the past five years in both digital signal processing hardware (DSPs) and in image-analysis software have made machine vision systems both highly capable and affordable. Prices of under $5,000 per unitcomplete with camera, illumination, processing hardware, and intelligent softwareare typical. Vision systems, for example, now are used to analyze all of the labels affixed to a product or to its printed box and adjusts, in real-time, the process to ensure that every item produced is to spec. Prior to the availability of affordable machine vision systems, human inspectors were stationed up and down each production line to "baby sit" multimillion dollar production machines. These are difficult and mind-numbing tasks for humans to accomplish effectively, but they are perfect jobs for machine vision systems, which never blink and which never get bored.

Dr. Bob goes on to describe the challenges that machine vision had to overcome to be useful in the "real world":

It is quite easy for a person to determine if the cap on a toothpaste tube is screwed down properly, but it is very difficult to design image-analysis software that will work reliably under all possible factory-floor conditions and product variations. For example, it might be relatively easy to write software that could reliably determine the position of a red cap on a white tube, but it would be far more difficult to create software that could determine the exact position of a white cap on a white tube.And, when you add to that the complexity of "real-world" manufacturingthe fact that the tubes are moving by at rate of 10 per second, that they are vibrating, that they don't always appear the same (for example, some tubes might be dented), that they are not always in the same position, that the designs of the tubes and caps are often changed by the marketing department (often, without telling the production department!)you realize the technical challenges that designers of machine vision systems faced. And, adding to those technical challenges, machine vision systems also had to be low in cost and easy to use in order to achieve widespread acceptance. Given those challenges, is it easy to see why there are so few successful machine vision companies today?

The good news: With modern CPU horsepower and years of work improving the associated algorithms, machine vision can finally solve these complex challenges.

Five or so years ago, the machine vision paradigm was to communicate the raw analog signals produced by standard video cameras through coax cables back to a remote computer system that performed the analysis of all the images. These early machine vision systems were too expensive and too difficult to use for all but the most demanding applications. But, because of the increasing power and decreasing cost and size of digital signal processing chips (DSPs), modern machine vision systems can now be about the same size as a cell phone, and contain a camera, illumination, vision software, and image-processing hardware. And, because they are low in cost and compact, these "vision sensors" can be placed at every point of the production line where value is added. They automatically snap images of each product moving by, and then they communicate pre-processed results (not the images) back to the factory control systems, which take the necessary corrective actions.

Much of the data gathering and processing challenges are akin to those of RFIDtons of packets of information snippets that must be stored (cached) and processed using complex-event processing models (e.g., mis-cappings on toothpaste is increasing while another system is detecting mis-feeds in the cap-loading line). And, as vision systems become more and more prevalent, more real-time information will be made available for correlation with a wider set of processes in the manufacturing complex.

"For example, today, an executive at a paper products manufacturing company can now sit in his office and literally 'watch' 10 different manufacturing plants around the globe and compare their production results," explains Dr. Bob. "In today's world, it is not good enough to wait for the end-of-week manufacturing report. A business executive needs to know the exact status (including quality metrics) of all of his production facilities so that he can optimize them." In the new world of higher intercompany tie-ins, the data and images could also be made available to both suppliers and customers, in real time.

Imaging is data-intensive and will strain back-end existing systems. Thankfully, the costs of key enabling technologies (disk space, processing power, and 10Gb Ethernet networking gear) are all trending downward. As a result, image data gathered from a manufacturing operation can now be "mined" more economically. Engineers can sift through millions of images and look for trends, correlations, or deviations against time, or against established specifications for components that are sourced from outside suppliers. Historically, the manufacturing segment has never had this kind of data available. Moving forward, we can predict that, because of their affordability, these systems will be used, and that data correlations will emerge that have not yet been anticipated. As in the usage of Inescapable Data technologies in the medical and commercial segments, the mining of image data will have extraordinary value to manufacturers as well.