• Font Size:
  • A
  • A
  • A

Feature Articles

Recycling Gets “Flakey” With Machine Vision

by Winn Hardin, Contributing Editor - AIA

Just a few years ago, there was money to be made in recycled materials. Paper made in the United States was recycled and returned to China to be used as packing materials for electronics and household goods during return trips across the Pacific. Plastic bottles for detergents made from colored HDPE plastic would be sorted by color, shredded and returned to bottlers for use as raw material.

Today, consumers are buying fewer goods, which mean less demand for commodities, including recycled plastic and paper. In 2009, towns like Branford, Connecticut had to start paying waste management companies to haul away recycled goods instead of collecting revenue. This has slowed the use of machine vision in sorting recycled materials. But ever the adaptive technology, machine vision has simply moved from bottle sorting to flake sorting, while using new standards and technologies such as GigE Vision to improve services in agriculture, postal and other sorting applications.

A Premium on Color

“In recent years, there was a premium on sorting plastic bottles by color,” explains Robert McConnell, President of machine vision software supplier, WAY-2C (Hingham, Massachusetts). “These applications really showed of WAY-2C’s strength because the HDPE bottles were crushed, irregular in shape, had multicolored labels and varying sizes - all going through the sorting machine at high speeds. We had to do a very fast color decision. The same was true of PET beverage bottles where European recyclers needed to separate the clear plastic PET bottles from light blue Evian bottles, for example. Today, we see more recyclers asking for machine vision systems that can sort the plastic flake, after the bottles have been shredded, rather than the bottles themselves.” 

WAY-2C software is based on a combination of information theory and special algorithms that simplify training machine vision systems for complex color applications while retaining the ability to make fast, precise decisions. “Most machine vision software techniques, not all of them, assume an object is a single color,” says McConnell. “People will say a phone book is yellow, but it’s actually yellow, blue, black, and other colors. Most color systems start by defining as few colors as possible. We train our software in contextual mode, taking into consideration the conveyor and other background objects, and assume that there are an infinite number of colors involved in the application. It’s a more difficult assumption, but works better in complex color environments. To accomplish this, we focus more on classes of objects based on unique color distributions in regions rather than a single color, pixel, or the isolated object.”

Defining classes - and in particular what colors best represent those classes - is often where the ‘magic’ of the machine vision integrator comes into play. In some cases, WAY-2C suggests that clients exploring new color sorting applications should consider testing their application with a hyperspectral or multispectral camera. These cameras create “image cubes” that essentially build stacked ‘cubes’ of images that show the same object over and over when viewed from different, narrow bands if the visible (or UV and infrared) spectrum. This can help system designers to figure out exactly what bands work best to find the defective elements in the image. Then the final system can be designed using cost-effective filters, rather than the expensive and sensitive hyperspectral camera. “WAY-2C normally works with three channels, each representing a different color,” says McConnell. “But it doesn’t care what those ‘colors’ are.”

A good example of the lessons learned from high speed recycling applications being applied to other industrial applications, McConnell says, is the growing use of machine vision in agricultural applications. “While for confidentiality reasons I can’t disclose the exact product, one of our customers wanted to sort items based on the amount of four different components present. Using our system they reported successful sorting at rates up to 15 per second, classifying each object based on weighting routines that are built into WAY-2C.”

The approach isn’t without computational demands, however. McConnell says that performing precision color inspection routines for high speed sorting applications typically requires multicore processors so that you can assign individual cores to different tasks, such as image processing, user interface, and network communication.

Gigabit Decisions On the Fly

Another way to handle data-intensive, high-speed sorting applications is the use of networked video and control architectures facilitated by the latest version of the GigE Vision standard, revision 1.2.

Machine vision postal sorting application, courtesy of Pleora Technologies Inc.“We’ve spent the past 10 years refining our technology to address concerns of latency and reliability with Ethernet networks,” explains John Phillips, Senior Product Manager at Pleora Technologies (Ontario, Canada). Pleora specializes in GigE Vision compliant video network solutions based on its iPORT and vDisplay lines of IP engines and eBUS-PureGEV package of high-performance PC drivers and SDK. “As a platform for high-performance industrial networking, there is nothing better in terms of flexibility, scalability, reach, and cost than GigE Vision.”

“When GigE first came out for industrial applications in 2003, it was viewed primarily as a point-to-point replacement that offered extended distance at lower cost compared to Camera Link and LVDS,” continued Phillips. “In the past few years, things have evolved. We're seeing increased market acceptance of GigE as a platform for real-time video networks based on switched architectures and multicasting. These architectures bring a whole new dimension to applications while lowering infrastructure costs; and they are now compliant with GigE Vision. In Version 1.2, the standard was expanded to include new types of video network elements, including receivers that display video directly on standard high-definition monitors without the need for PCs with video cards, as well as video servers, video processing units, data management entities and network-controlled devices. And with multicast capability, or the ability to address the same package to multiple network devices, GigE Vision devices are able to send data to multiple processing and display elements without dramatically increasing bandwidth."

In one recent example, Pleora’s iPORT IP engine video transmitter was integrated along with its vDisplay video receiver into a postal sorting application. As envelopes pass along the chute, cameras capture an image of the address at very high speeds. The iPORT IP engine sends the images out over a GigE Vision compliant network and acts as a programmable logic controller (PLC) for return commands from the analysis PC. By using the iPORT IP engine as a local PLC, Pleora can provide additional actuator control throughout the sorting machine without adding third party PLCs. “Pleora’s iPORT IP engines are equipped with PLC technology that allows users to synchronize, trigger, and control the operation in even the most complex, precisely timed vision applications,” adds Pleora’s Phillips. “Accessible from the eBUS-PureGEV SDK, Pleora’s PLC combines I/Os, timestamps, delayers, rescalers, counters, and other advanced features into one powerful, highly flexible tool.”

“In the case of this postal sorter, the camera is much closer to the envelope actuator than the analysis PC. By using the PLC within an iPORT IP engine and the GigE network to control the ejector, the PC and other processing elements can be removed from the sorting and conveying system. This allows the integrator to locate the control and analysis PC and servers to protect IT closets while reducing the need for additional cabling. The IP engine communicates to other network elements across a single GigE cable with a short cable run from the camera to the actuator. By adding a vDisplay receiver, operators can see the images from the cameras to make sure everything is running as it should be without adding a PC with display card or increasing bandwidth requirements. GigE Vision’s multicast capability means that the iPORT IP engine can address packetized data to multiple network devices without increasing the required bandwidth.”

Sorting applications, whether it be recycled bottles or letters to Santa, are tailor made for distributed network solutions like GigE Vision, concludes Phillips, where companies want to add the power of machine vision without extensive network infrastructure and hardware add-ons. “With GigE Vision now well established as a video delivery platform, we're finding customers want to move beyond basic camera-to-PC connections and take advantage of real-time switched video networking.”



Siva Selvaraj   (February 5, 2013 01:53 PM)

I would like to know more about the vision processing in recycling operations. Please call me at 9254370868.


Leave a Comment:

All fields are required, but only your name and comment will be visible (email addresses are kept confidential). Comments are moderated and will not appear immediately. Please no link dropping, no keywords or domains as names; do not spam, and please do not advertise.

First Name: *
Last Name: *
Your Email: *
Your Comment:
Please check the box below and respond as instructed.

Search AIA:

Browse by Products:

Browse by Company Type: