IEN: What issues confront machine vision? How can they be resolved?
Vargas: There is a clear trend of increasing complexity for vision applications. Original systems were only capable of binary (monochrome), low-resolution camera, and a PC as the controller (their original platform) with some costumed program. Current systems are compact, standard resolution is 480 x 512 pixels, capable of gray-scale detection (255 gray-scale), and even multiple color detection with advanced algorithms, such as pattern detection, optical character recognition (OCR Type uses Neuro-Net Algorithms), flaw detection (Scratch and dent detection -- AG50 series), CONTOUR Matching (360 Degree -- Object Outline Recognition).
This new line of low-end vision systems is capable of performing faster and better inspections which in the recent past were only possible by human operators. The new generation of products is constantly upgraded with additional functions and special algorithms to provide quantifiable data and reliable inspections.
In my opinion future systems will be required to detect physical attributes in a similar way as human operators do, focusing on features such as color, size, shape, and distinguishing markings as part of a single and specific product, rather than performing multiple and independent inspections. The usage of Artificial Intelligence software within compact CPU platforms and the addition of mega-pixel resolution cameras will enable these systems to distinguish complex patterns and features for multiple products, while reducing the hardware cost and remaining affordable, easier to use, faster, and more reliable, in order to justify their usage in an industrial environment.
IEN: Has the use of vision systems spread beyond traditional end-use markets, and if so, why?
Vargas: Current and newer vision systems are rapidly moving into areas usually considered human inspection domain. Specifically in the following industries:
- OCR for Food and Pharmaceutical Industries: Most machine vision systems have only limited accuracy and hence marginal success performing optical character recognition. Medium to low end vision systems (under $20,000 price range) will provide less than 93% accuracy detecting characters, especially characters created using a dot matrix printing method (labels on pills packaging), or metal stamped characters (part stamping). Machine vision systems with advanced Neuro-net algorithms such as our A230 OCR series and soon to be released AG50 are already capable of detecting such characters accurately and reliably.
- Flaw Detection for Automotive Industries: Our soon to be released (3rd quarter 2003) AG50 mega-pixel camera system uses advanced algorithms to detect flaws within low or poor contrast surfaces such as scratches, rust spots, dents and protrusions within shiny metallic surfaces. Such complex inspections are traditionally delegated to the human operator, due to the innate human capability to recognize patterns. This new system uses its Neuro-net algorithms to generate a quantitative, reliable, and non-subjective inspection of such materials and surfaces.
- Part Tracking for Automotive Industries: Traditional automotive assembly machines required an operator to function as a human Pokka Yoke* machine. (*Japanese term means checking for presence or absence of certain conditions within a specific part.) Our new and more robust systems such as the AX30 (Color) and AG50 (Mega-pixel) provide advanced "CONTOUR" algorithms capable of tracking the "outline" of specific parts within 360 degrees, even if the object is partially obscured by overlapping parts, while checking for additional conditions on these parts. This new technology enables our vision systems to combine these new tools with more traditional inspection such as OCR (optical character recognition), feature detection, and color extraction to provide a smart and complete inspection package right out of the box.
These types of advanced systems were only available for high-end applications with dedicated software on PC-based systems. Now they will be available not only at a more reasonable cost, but additionally will provide standalone capabilities inside a rugged and industrial hardware platform.
IEN: Will wireless play a role in sensing? In machine vision?
Vargas: For the majority of industrial applications wireless networks are not yet accepted or in some instances not welcome -- especially in applications where AC drives (inverters) or multiple drive systems are used. Such devices (electric motors in general) generate electromagnetic noise throughout the entire range of harmonics, creating a phenomenal amount of interference.
In my opinion, I have no doubt that wireless networks will eventually become the standard method of communication for office and home applications within the very near future. And all of their inherent problems, such as interference susceptibility, network security, and reliability issues, will be resolved. However, for industrial applications, utmost reliability and bulletproof network security are musts. I think the use of wireless networks and protocols will take a few years longer before they are accepted and used.
IEN: Can sensing be part of a lean/flexible manufacturing solution? How about machine vision?
Vargas: Machine vision''s main purpose in any manufacturing environment is quite clear. It provides reliable, accurate, and automatic visual inspection of products on the production line wherever it is used. It reduces human labor (great for the bottom line), reduces tedious inspection tasks (great for human operators), and increases quality (great for consumer satisfaction and sales).
Traditionally the usage of a vision system meant expensive computer equipment, specialized lighting sources, and a battery of system integrators to perform the programming and configuration. Within the new economic environment the explosion in technology had provided a ripe environment for affordable vision systems, with standalone software (no need of PCs) and faster and higher resolution cameras.
Within the last 10 years machine vision trends have been lowered from upwards of $50K to under $5K dollars (pricing and sales data can be obtained/confirmed from Advanced Imaging Association 5 year vision reports), and prices are still trending downward. Due to the fierce competition among machine vision suppliers, prices are still dropping, and their capabilities are improving. Color extraction, object tracking, optical character recognition, and mega-pixel cameras are now available.
While their prices are dropping significantly, machine vision systems are reducing labor costs and improving product quality while gathering production data for analysis. Such data can be collected to improve the manufacturing process and increase product quality significantly.