Today we are hearing the term ‘‘vision sensor’‘ being used with reckless abandon. Is a ‘‘vision sensor’‘ a machine vision system? Well, yes and no. In many respects a vision sensor utilizes the same technology infrastructure as a machine vision system – camera, lighting, optics, compute power – and typically performs a control function (quality control, process control, machine control, robot control) based on analyzing image-based data. On the other hand, the functionality of a vision sensor is generally limited to a single or maybe a couple of generic machine vision applications – sensing presence/absence, alignment, etc.
Typically, a machine vision system has greater configurability – can handle a broad range of applications often utilizing generic functions but, for many applications, in combinations – find, OCR, 2D symbol reading, presence, pattern recognition, etc. In other words, if an application requires more configurability, a simple vision sensor will not work.
Seems that machine vision is more of a continuum than discrete and distinct segments. Vision sensors are on one end of the continuum with proprietary, application-specific systems on the other end. As much as anything the distinction between ‘‘system’‘ and ‘‘sensor’‘ may be in degree of configurability. Most sensors have little or no configurability while configurable machine vision products are just that and these include smart cameras, frame-grabber-based PC systems, or PC-based systems with intelligent boards, PC with FireWire, USB, Ethernet camera connectivity, development systems with proprietary designs (chip sets, special boards with DSPs. FPGAs, ASICS, etc.). Increasingly it's the software that distinguishes one system from another regardless of the hardware implementation. In some cases systems are furnished with general-purpose software, some cases with software for specific functions (OCR, alignment, etc.) or combinations of the two.
With all the advances in computer technology the vision industry is no longer easily segmented. Notwithstanding, one has to recognize that there is also a continuum in terms of application difficulty, so while in many cases these products compete with each other, in most cases the application difficulty will dictate specific segments of the products that will compete with each other.
What is most interesting is to read the literature with the suggested range of applications that can be addressed by a simple vision sensor. Some of the earliest pieces of product literature I wrote back 25 years ago suggested many of the same applications. The big difference is that today the vision sensor can perform those applications for under $2K, whereas 25 years ago the final installation of the machine vision systems would cost at least $25K.
Whereas at $25K+ justifying equipping multiple lines was a formidable task, today at under $2K using a vision sensor to monitor each value adding function along a production line can be easily justified. In many cases capturing just one condition that will lead to customer dissatisfaction and field returns, warranty costs, etc. will result in immediate payback for purchasing the vision sensor.
What does a vision sensor look like? Well there are two distinct configurations identified with the term vision sensor. One configuration is that of a self-contained unit in a size that generally fits in one’s hand. Within the unit one finds the image-based sensor, optics and compute power in the form of a microprocessor, memory, DSP or FPGA or some combination of these as well as the input/output connectivity. Lighting may or may not be an integral part of the unit. In the second configuration, the camera is tethered to the vision computer box, so to speak, which houses the compute power and associated electronics and input/output connectivity. In both implementations, the man-machine interface for setting up the vision sensor is straightforward and makes the underlying machine vision technology totally transparent to the end-user.
To gain additional insights on vision sensors from the vendors’ perspective, a request was submitted to all known suppliers of vision sensors with specific questions. The answers provided in a ‘‘round-robin’‘ format follows. The following companies provided answers:
1. How would you describe your vision sensor?
[Jon Hickman – Banner Engineering] Banner Engineering Corp. offers a rapidly growing vision sensor line under our PresencePLUS? family name. The full-featured PresencePLUS Pro offers a variety of vision tools in a single package, for under $2000. The four PresencePLUS P4 models include two low-cost sensors that use a single tool ($995, with 128 x 100 resolution) and two high-resolution sensors (1280 x 1024 megapixel). All PresencePLUS sensors set up with the same easy-to-use Windows? interface—a graphical ‘‘wizard’‘ that walks the user through configuring an inspection.
[John Keating - Cognex] We don’t use the term vision to describe Checker. Checker is a high-end sensor that is a better solution for applications that would require multiple photoelectric sensors, or for detecting features that photoelectric sensors cannot detect.
[Mark Sippel – Cognex] In-Sight vision sensors are defined as ‘‘any networked image capture and processing device that is user configured for general-purpose machine vision processing tasks using self-contained imaging and processing with input/output capability and a non-programmatic GUI’‘.
[Robert Lee – Omron] Application Specific Machine Vision Sensors, General Purpose Smart Sensors, Entry Level Vision Controllers.
[Wayne Meyer – SICK] The SICK Intelligent Camera Sensor (ICS) series uses a CMOS array receiver, with up to 520 by 520 pixels. The sensor is designed to typically incorporate the image sensor, integrated lighting and lensing as a single component. The ICS family of vision sensors offers larger fields of view - up to 3.1 x 3.1 inches with integrated lighting, or 6.3 x 6.3 inches with external lighting. All teach windows and search windows are fully adjustable. The ICS 009 is also available with a C-Mount lens, which allows the user to define the fields of view, thus making the ICS Vision Sensor more user friendly.
[Endre Toth – Vision Components] Vision Components currently offers more than 20 different models in three product groups:
VC smart cameras present a unique balance of size, performance, power consumption and most importantly, cost. They are all self-contained units in a robust industrial housing with integrated on-board DSP image processing, designed to capture images, extract features from the image array and output data. The smart cameras are incredible compression units considering that they often output one bit, or a few bytes of data from the millions of pixels they capture, requiring narrow bandwidth to transfer the data.
With the board-mounted smart cameras VC occupies an unmatched position in the market place. As we also work with system integrators and sensor suppliers, it is not always possible to identify our products as VC vision sensors. It might well be the case that you are already using VC vision sensors without knowing it.
2. Can you provide a one or two paragraph general description of the underlying principles associated with your vision sensor implementation? Image processing/segmentation principles? And image analysis principles?
[John] The overriding principle of Checker is simplicity. Because of Checker’s speed, processing at 500 frames per second, Checker is able to detect a part without a photoelectric sensor, and can precisely determine when the part was detected. This allows Checker to have deterministic output timing, which makes Checker very simple to integrate on a production line.
Checker doesn’t have ‘‘vision tools’‘, Checker has sensors designed to detect the presence of parts and part features. Checker’s Part Finding and Inspection sensors were designed to look and feel like photoelectric sensors so the user would be immediately comfortable using them.
[Mark] In-Sight captures images based on user-set criteria, and processes these images with all of the advantages that DSP technology has to offer. In-Sight vision sensors capture framed image data varying in size from 640x480 monochrome or color to 1600x1200 pixels monochrome in size and can process the full-framed image or any percentage as determined by the user’s settings.
Each In-Sight is a network-enabled device that uses a vision spreadsheet GUI on a PC to configure machine vision measurement and processing tasks for general-purpose vision applications. These tasks include simple to complex gauging, inspection, guidance and identification capabilities. The number of tasks, image capture and I/O functionality configured determine the length of time for each job cycle. ‘‘Job’‘ files, as they are called, contain all of the necessary information for the In-Sight vision sensor to process the user configured vision tasks. Output from the job execution results can by made either digitally or by several data-based communication protocols.
[Wayne] The ICS is a unique vision sensor that uses a simple teach-in function to store the correct value of pixels in the image. Whether you use Pixel Sum, Minimum Pixel Sum, Area Evaluation or Shape Check, the setup process is simple. The algorithm is selected using a hand-held programming tool, the VSC, which can be detached upon commissioning of the ICS. An Area Gloss Detector (AGD) variant is designed for Area Gloss Detection, based on the minimum pixel sum algorithm. A Date Code Inspection (DCI) variant is designed for Date Code Inspection using a pixel sum algorithm.
[Endre] The VC Software Library tool kit includes highly optimized, high performance image processing, segmentation, feature extraction and analysis tools. Users can easily put together vision sensors by combining functions, such as: Contour8, Sobel operator, ff3 filter, AND, OR, MIN, MAX, look-up-table, average, pyramid, erode, dilate, correlation, histogram, image arithmetic and many others. Separate application-oriented library tool kits are offered for bar code reading, data matrix decoding on a one-time license.
[Jon] Banner has a history of leading advances in photoelectric sensing technology, which gives Banner the edge in developing two-dimensional vision imaging. We’ve developed proprietary algorithms for analyzing images quickly—for example, the PresencePLUS P4 EDGE can accurately detect more than 10,000 objects per minute. And Banner’s advanced knowledge of lighting techniques for creating contrast for a traditional single-pixel photoelectric gives us a tremendous advantage in devising lighting for a multiple pixel image.
3. Are the image processing and analysis tools accessible for modification or addition by the user? If so, describe, how.
[Wayne] Users can see the image on a 256 grayscale handheld (VSC), so they know exactly which image they want to detect. The VSC allows for easy teach-in and modifications by the user.
[Endre] Along with the smart camera hardware, customers purchase the cross-platform software development kit for a one-time license fee from Vision Components, which appears to be a distinctive arrangement in the industry. With the SDK the user can modify existing algorithms, directly control all hardware features, or create new analyses tools from scratch. The openness and flexibility of the SDK offers a large span for code optimization to reach the performance the user needs for an application.
[Jon] All PresencePLUS sensors let the user define what the sensor sees. A wide variety of lighting options and lens choices give the user the flexibility to mount the sensor where it is most convenient and, often, to light the inspection in a variety of ways. Each vision tool can be as large or small as required to see exactly and only the target. Each inspection includes as many tools as needed, including test tools that allow logical analysis within the sensor itself.
PresencePLUS sensors are designed for easy, cost-effective implementation on the factory floor, making possible inspections that do not justify a costly machine vision system. The PresencePLUS sensors offer a graphical setup with a suite of tools: location, normalized gray-scale correlation pattern matching, geometric feature extraction pattern matching, BLOB, average grayscale, edge and object. The inspections and output type are easily accessible for modification.
[Mark] Most of the vision tools used on an In-Sight vision sensor can be accessed to modify the tool’s properties based on the users needs. Most tools have several properties that affect things like inspection area size, number of pattern instances to search for, image parameters and even display options. These same tools can use data and positional references provided from other tools by providing X, Y or Theta position references to set the measurement area position of the particular tool and can also provide sample values from one tool to another.
4. Can you provide a one or two paragraph general description of your graphic user interface/man-machine interface? What interface steps need to be taken to set up an application?
[Wayne] Using the teach-in keys, users can easily store the image for inspection. The VSC is just slightly larger than a handheld PDA. It connects with a cable to the SICK ICS. Upon completion of the programming, the VSC is typically detached. Several sensors in a plant can be programmed using a single VSC. The teach-in process depends on the complexity of the inspection task. The AGD and DCI application-specific variants of the ICS have a simple 2-step teach-in process. The ICS sensor holds the parameters in its own on-board memory.
[Endre] Using the library tool kits, partners can develop graphical user interfaces. In fact third parties offer several different user interfaces, some of them are script based, spreadsheet based, menu-based and icon-based. In some implementations the user interface is running on a connected PC or PDA. In others the user interface itself is also running on the DSP of the smart camera without a PC. This solution takes full advantage of the SVGA output embedded in Standard and High End VC smart camera models. No PC is necessary even for setup.
[Jon] The software for all PresencePLUS sensors runs on all current versions of Microsoft Windows? and offers a simple 3- or 4-step process to set up an inspection. On the Set Up menu, the user mounts the sensor, focuses the lens, and sets light levels. Using the Tools menu, the user specifies the vision tools, measurement tool, and test tools for the inspection. At this point, most users can simply click Quick Teach to transfer the inspection from the computer to the sensor and click Start to launch the inspection. If desired, the software can monitor a running inspection, providing such data as total processing time, live video grabs from the sensor, individual tool feedback, and Pass/Fail counts.
[John] The interface is the key to Checker’s simplicity. The CheckMate setup software provides a step-by-step setup of a Checker job. CheckMate walks the user through 4 simple setup steps, and along the way, provides simple instructions on what to do in each step. The user never needs to go looking for a hidden menu or feature – everything the user needs to do is in the step-by-step process.
The CheckMate setup software was designed to ensure there are no ‘‘vision’‘ terms anywhere in Checker’s setup software – it uses terms and concepts that the customer already understands. The part finding and inspection sensors were also designed to be simple to use – they don’t have numbers or parameters to set. They have a simple output meter and a sensitivity adjustment similar to the gain of a photoelectric sensor.
[Mark] The In-Sight Explorer integrated configuration interface provides the means to easily connect cameras to the PC, configure the vision application using a vision spreadsheet interface, and even create an operator interface for the vision sensor within the same software package.
Once an In-Sight is connected to the PC, a user drags and drops the required vision tool into the spreadsheet. A property sheet opens allowing the user to configure the setting in the tool including things like the measurement region and tool tolerances using graphic overlays on the image or drop down menu settings. No programming of any kind is necessary to configure tools. The operator interface also can be created in the spreadsheet using drag-and-drop tools and standard cell formatting like standard Windows? programs to create a run-time operator interface with indicators and value readouts and inputs.
[Robert] Simple to use GUI/Drop down menu/Icon driven
5. How do you distinguish between a vision sensor and other configurable vision products, such as smart cameras and embedded vision processors?
[Endre] The features of a vision sensor:
Often vision sensors are smart camera-based. A smart camera combined with application-specific software creates a vision sensor. Some vision sensor solutions provide a single function, such as differential image sensor, diameter sensor, roundness sensor, color sensor, OCR reader, date code verifier and so on. Others offer a set of basic functions, such as pattern matching, edge detection and so on, which can be combined at the time of configuration.
[Jon] A vision sensor analyzes and interprets data from an entire image scene, rather than just a single point. The data analysis occurs in the sensor itself. Sophisticated vision sensors, such as those in Banner’s PresencePLUS line, can detect more than a million pixels, so they can capture a large area in great detail. A vision system also analyzes and interprets detailed data from an entire image scene, but uses a computer with complicated custom software to analyze the data.
[John] Checker is different from both vision sensors and configurable vision products. Checker inspects for the presence of parts and part features and provides a simple, deterministic, pass/fail result through opto-isolated outputs. There is no data to interpret or to communicate, and there are no ‘‘vision tools’‘ to use.
[Mark] We do not make distinctions between them if they fit the criteria mentioned above in question 1.
[Robert] By the algorithm it employs, the application, and the specific results derived from the sensor itself.
[Wayne] The vision sensor provides a simple discrete output and is typically set up very quickly. Ideally, a vision sensor will be like a basic sensor (photoelectric) that incorporates its own lighting and lenses, without having to assemble a ‘‘system.’‘ There are a limited number of algorithms, or settings, to choose from. Smart cameras typically require much more demanding programming. While the programming is often point-and-click Windows?-based, it typically requires a higher level of training or learning.
6. Why are vision sensors so inexpensive?
[Jon] In the case of Banner, we manufacture the products we design, enabling us to control costs. We know how to design for manufacturability. Also, we offer different vision sensors to fit a range of applications, not every possible application. The user selects and pays for only what is needed to reliably solve their problem. Expensive machine vision systems tend to combine a variety of tools, uses, and in some cases, multiple pieces of software in a single, often unwieldy package.
[Mark] Product cost is a reflection of both the lower cost of manufacturing and a limited degree of functionality and purpose. In-Sight vision sensors are made specifically for their task. They are proprietary and cannot physically be expanded. This limits costs in manufacturing and functionality and provides a consistent platform to support.
[Robert] Market segmentation.
[Wayne] Discrete outputs and limited programming options minimize the additional internal componentry necessary, which can cut down on costs. Also, lower costs for CMOS receivers due to the advancement of technology and electronic trends -- digital cameras and cell phones - have helped drive down the costs of the components.
7. What are vision sensors a substitute for? Or are they addressing applications not heretofore addressable with existing sensors?
[John] Checker is a substitute for applications that would have required multiple photoelectric sensors and PLC logic. Checker is a better and more cost effective solution compared to the cost and complexity of wiring up and maintaining multiple photoelectric sensors. Since Checker can detect parts and provides deterministic output timing, Checker is also able to solve applications that no other sensor can solve, such as using a pattern as a registration mark.
[Mark] In the case of In-Sight vision sensors, there are some application spaces where they are replacing PC-based products mainly because these applications can be better served with the functionality and simplicity provided by this platform. But in many cases, vision sensors are addressing new application spaces that traditionally did not use vision. These spaces typically relied on human intervention or on the known defect values associated with the production equipment used. They are also being used where several combinations of technology may have been used in combination to accomplish the same task, but with a high degree of unreliability.
[Wayne] Vision sensors often replace multiple sensor groupings with a single device. They also perform more detailed inspections than are possible with standard photoeyes -- like inspecting a larger area (field of view) than a single sensor. In addition, some vision sensors are used in place of Smart Cameras or vision systems that would be ‘‘over-qualified’‘ for the application. An example of this might be the inspection of a date code with the SICK DCI: instead of a complex vision system that utilizes some kind of OCR/OCV, most of the print-verification requirements of the inspection is to merely make sure that something is printed, and not necessarily that the particular characters are present (most often the printer is not going to change the letters, but instead have problems with print quality, such as low ink or over spray and smudges).
[Endre] Vision sensors unite the advantages of image processing and traditional opto-sensors. With the combination of the two the limited application range of opto-sensors has significantly expanded.
[Jon] Some PresencePLUS vision sensors replace older, more expensive machine vision systems, where the complexity of the product was overkill for the basic application. And PresencePLUS is finding its way into applications that never could be solved with sensors before. The ability to cover a large two-dimensional area at the price of a sensor has opened a whole new world of opportunities.
8. Are your vision sensors general-purpose or application-specific?
[Wayne] Both. The ICS is considered a general-purpose sensor and the AGD and DCI are for specific applications.
[Jon] It depends on the product. The PresencePLUS P4 BCR is specific to reading 2D and 1D bar codes. Other sensors in the PresencePLUS family can be used across a wide variety of applications when coupled with the proper light and lens.
[John] General purpose is too broad a category for Checker, which could be considered multi-purpose. Checker detects the presence or absence of parts and part features, addressing a wide variety of applications across all industries.
[Mark] In-Sight vision sensors are generally considered general-purpose, but we do make an ID variant specifically for 1D and 2D barcode reading.
9. If general-purpose, what have been some applications for which your vision sensors have been found particularly well suited?
[Wayne] The ICS from SICK can be used for the tasks that have been developed into the specific applications: area gloss and date code inspection. In addition, inspection for proper surface treatment in die cast parts processing (looking to see if there are holes, threads, surface characteristics, etc.). The ICS can be used to verify complete assembly of small parts in medical devices.
[Endre] Applications lie on a very broad scale in many different industries, as customers become more intuitive every day inventing more and more applications, some times very surprising ones. The most common areas of application include integration into industrial production processes, quality assurance or safety equipment, measurement technology, filling control, mechanical engineering, motion analyses, optical character recognition, biometric access control, robot guidance and detecting barcodes or 2D matrix codes.
[Jon] Verification, Presence/absence, Gauging, Measuring, Orientation, Flaw Detection, and Sorting to name just a few.
[John] Checker is designed for presence absence applications, and has been applied in a wide variety of applications in many industries. In the automotive market, Checker has solved many error proofing applications throughout the automotive manufacturing process including verifying the presence of piston rings, weld nuts, threads in holes, correct buttons, clips on seats and many more.
In the food and beverage industry, Checker has replaced many photoelectric sensor applications – especially where line changeover is required. Some applications include cap presence, safety seal presence, upside down bottle detection, case packing verification, inkjet code print presence, label presence, correct label presence, insert presence and many more.
In the packaging world, Checker’s deterministic output timing was the key to being used on shrink wrap machines to accurately detect a pattern so the shrink wrap can be precisely cut – without requiring a registration mark.
[Mark] In-Sight vision sensors are very effective in solving gauging applications in the automotive markets, label inspections in consumer, pharmaceutical and food and beverage industries, content presence/absence and fill inspection in food and beverage, ID applications for both 1D and 2D barcode and OCV applications in
by Nello Zuech, Contributing Editor - AIA Posted 09/08/2005