- This article is filed under:
- » View All
The Case for Frame Grabbers . . . . by Nello Zuech, Contributing Editor
by Nello Zuech, Contributing Editor - AIA Posted 12/07/2005
I can remember working for EMR Photoelectric back in 1973 and we were developing machine vision systems – although they were not called that then. A major challenge was taking analog signals and converting them to digital signals and then storing the image so one could operate on it in some application-specific manor. Our very creative engineer came up with what subsequently became called a frame grabber. It incorporated an A/D and sufficient memory to capture a 256 x 256, 4-bit image. Little did we know that he actually had developed a board-level product that would have been of interest to other companies embarking on processing images. The board itself was embedded in an OCR system and subsequently in a system that made dimensional measurements on bomb fuses.
Subsequently, many companies emerged to provide frame grabber functionality. Most often the basic versions of these boards include an A/D, a look-up-table, memory and D/As for display purposes. Many companies offering basic frame grabbers have versions that incorporate compute power, originally in the form of specific board-level hardware and most recently in the form of application-specific integrated circuits, microprocessors, digital signal processors and/or FPGAs. In any event, these board-level products have come a long way, so to speak.
As the PC advanced, using a frame grabber with a PC was an approach to eliminating the requirement for proprietary machine vision engine hardware designs. The PC with the frame grabber became a vision engine when combined with machine vision software. Today the simple, basic frame grabber is fighting for its life as advances in processor speeds in combination with digital cameras and standard interconnectivity techniques (IEEE-1394, USB, Camera Link®, Ethernet) now make it possible to stream the image data from the camera directly to a PC and the PC with appropriate machine vision software becomes the vision engine, without the need for complementary boards.
There are still many applications where this approach will not have sufficient image processing robustness or be able to keep up with the speed. Then there is still a requirement to offload image-processing functions onto hardware in the form of a frame grabber with compute power or an intelligent image processing board.
To gain some insights into the current state-of-the-art of frame grabbers and related image processing and manipulating boards that plug into PCs, an email was submitted to all known suppliers of frame grabbers with specific questions. The answers provided in a ‘‘round-robin’‘ format follows. The following provided answers:
- Bashar Mashal, Product Marketing Manager for PC Vision, Cognex Corporation
- Bruce Shulman, Director of Business Development, Cradle Technologies
- Philip Colet, Vice President, Sales and Marketing, DALSA Corporation
- Andrew Sharpe, President, IO Industries
- Stephane Francois, Executive Vice President, Leutron Vision
1. Can you provide a brief general description of your line of frame grabbers specifically targeted at machine vision applications? What specifically differentiates the respective models that you offer?
[Bashar Mashal - Cognex] Cognex is the world's leading supplier of machine vision systems, which includes a full line of frame grabbers. The MVS-8000 Product Family includes multiple frame grabber product lines designed to meet the needs of a wide range of applications where different levels of cost and functionality are required. The MVS-8000 family offers frame grabber and embedded vision hardware configurations, a comprehensive library of MMX-accelerated vision software tools, as well as a choice of form factors and development environments.
The Cognex MVS-8600™ is a family of digital frame grabbers that support the Camera Link communications protocol. These high-speed frame grabbers can be used with both area scan and line scan Camera Link compatible cameras and are ideally suited for high-speed and high-resolution machine vision applications that require very fast processing of large amounts of image data. The MVS-8600 frame grabbers utilize standard Camera Link cameras and cables to simplify and lower the cost of system integration. The MVS-8601 is a single-channel frame grabber; the MVS-8602 is a dual-channel frame grabber.
The Cognex MVS-8500™ family of analog frame grabbers supports multiple high-speed analog cameras that utilize the latest progressive scan CCD sensor technology. The MVS-8500 frame grabbers are also backward compatible with standard speed analog interlaced and progressive scan cameras. The MVS-8501 supports up to 4 multiplexed cameras; the MVS-8504 supports up to 4 independent cameras.
The Cognex MVS-8100L™ is a high-performance, low-cost frame grabber. The MVS-8100L is designed to capture and transfer high quality grey-scale images from up to four multiplexed RS-170 or CCIR cameras with image formats of up to 640 x 480 (RS-170) and 760 x 574(CCIR).
The Cognex MVS-8100C™ is a high-performance machine vision frame grabber for applications that require a color image display. It provides high-speed image transfer to the PC for processing and color display, enabling fast vision application operation and a single-monitor operator interface. The MVS-8100C frame grabber offers dual window display where one window presents frozen or live color image display and the other shows the monochrome image used for analysis.
The Cognex MVS-8120™ is a high-performance, flexible acquisition PCI frame grabber that combines MMX-optimized PC-based processing and Cognex's comprehensive library of vision tools to provide a cost-effective, high performance vision solution. Designed with modular image acquisition architecture, the MVS-8120 supports a wide range of camera types including RS-170, CCIR, large format digital, and line scan cameras.
[Bruce Shulman – Cradle Technologies] Grouper is a 16 Channel NTSC/PAL ‘‘Intelligent’‘ Frame Grabber with 16 channels of NTSC/PAL Decode AND 11 GFLOPS MDSP Processor from Cradle
[Philip Colet – Dalsa Corporation] DALSA offers the X64 series of frame grabbers, specifically designed for high-end industrial applications. There are three primary models of the X64 series of products, which are designed along the lines of the camera interface. Therefore, the series comes in Camera Link, LVDS and analog models. Of course there are numerous variations even within one camera interface technology.
The X64 series delivers what DALSA calls – ‘‘Trigger to Image Reliability’‘, or TTIR for short. TTIR is a combination of hardware and software technology, working in tandem to ensure the reliability of the acquisition process, the recognition that sometimes things go wrong and the ability of application software to recover when things do go wrong.
[Andrew Sharpe – IO Industries] Our frame grabbers are designed specifically for video recording directly to hard disk. This is useful in machine vision applications where images captured during diagnostic testing of a machine or a process must be archived for off-line analysis. A special hardware feature on our frame grabbers allows video to be recorded at higher data rates than is possible when using a conventional frame grabber. This design innovation also makes it possible to easily record multiple cameras with a single computer system.
IO Industries frame grabbers:
- DVR Express CL160 - Maximum recording speed to hard disk is 135 MByte/sec from a Camera Link or LVDS video source.
- DVR Express CLFC-Full - Maximum recording speed to hard disk is 810 MByte/sec from a Camera Link video source.
- DVR Express CLFC-Dual Base - Maximum recording speed to hard disk is 243 MByte/sec from two Camera Link video sources.
[Stephane Francois – Leutron Vision] Leutron has been mainly a frame grabber manufacturer for over 25 years. For Camera Link, we have 14 different models (1 or 2 connectors, onboard processing, PCI or PCIe or PMC). In analog, we offer a choice of PMC, PCI or CompactPCI, for color or monochrome cameras with complex acquisition mode support.
2. How would you describe your frame grabber? What specific functions does it include?
[Bruce] Complete development environment and Board Support Package to allow easy porting and integration of customer’s algorithms -- existing libraries include compression and video feature extraction (extra charge). There are also Audio inputs.
[Philip] The X64 series is targeted at high-end machine vision applications and as such are designed to interface to very fast cameras. Camera and sensor designers are very concerned with sensor characteristics including speed and noise, two orthogonal variables. These camera vendors are not so concerned with what happens with the image data downstream from the camera and the output from many high-end cameras is multi-segmented data, which can appear to be mirror image, upside down or a combination of the two.
The X64 provides the ability to transform the multi-segmented data into a coherent image of a specified resolution and bit depth. This transformation takes place entirely on the frame grabber, relieving the host computer of the laborious task of reassembling data on a pixel-by-pixel basis.
Additionally the X64 provides on-board utilities for performing a fixed pattern noise correction, flat field correction and a Bayer conversion from color cameras.
[Andrew] Our frame grabbers are designed to be a video recording subsystem. The computer provides power, starts and stops video recording operations, but is not involved in the underlying operation of writing data to hard disk. An onboard disk controller chip enables our frame grabbers to write data directly to a set of hard disks connected to it. This differs from recording systems, which use a conventional frame grabber in that those systems rely on software running on the host computer to store video data to hard disks managed by the host computer.
By relieving the computer of the task of storing video data, we are able to record at higher data rates and from more cameras using a single computer system than is possible when using a conventional frame grabber.
[Stephane] High performance, cost competitive, feature rich.
- Choice of PC interface (PCI 64bit/66MHz, PCIe x1 and x4, PMC)
- 85MHz CL support standard
- Real independent dual base support
- Native support for Windows, Linux and VxWorks
- Intelligent Frame buffer (tap reordering, buffer for transfer)
- Extensive Line scan camera support (operation modes and encoder support)
- Onboard HW 5x5 non-linear bayer interpolation
- Trigger and Strobe control
- User I/O
[Bashar] In addition to the above-mentioned specifications, all Cognex frame grabbers in the MVS-8000 family provide onboard I/O. Depending on the model - triggers, strobes, encoder inputs, TTL and Optically Isolated inputs and outputs can be available onboard. Moreover, all of the above frame grabbers conform to the PCI specification and provide minimal latency between image acquisition and transfer to PC memory. Finally, MVS-8000 frame grabbers combine MMX-optimized PC-based processing and Cognex's comprehensive library of vision tools to provide a cost-effective, high performance vision solution.
3. Do you offer machine vision software for your frame grabbers? If so, please describe that software. If not, do you offer third party commercially available software or recommend such software, and, if so, whose?
[Philip] DALSA does provide machine vision software that works in conjunction with our frame grabbers. This software provides over 300 optimized image processing functions as well as image analysis functions for defect detection (blob analysis), OCR, Bar Code reading and geometrical search algorithms.
[Andrew] IO Industries Streams 5 application software provides control over the DVR Express frame grabber for video recording and playback operations. The software allows recording from multiple cameras, time stamping each video frame using computer time, IRIG-B time or GPS time. Support is provided to simultaneously record data streams from other sources including audio, GPS, navigational instruments, and data acquisition boards. An optional SDK allows C/C++ programmers to customize recording and analysis procedures. Image analysis can be performed in real-time or by post-processing.
[Stephane] We only supply SDK to control the boards. We only suggest some software: MVTec Halcon, MVTec ActivVisionTools, Neurocheck, VisionBlox...could work with all software.
[Bashar] Cognex is the world's leader in machine vision software and offers two development environments that enable customers to develop their applications.
The CVL® (Cognex Vision Library) Software Developer's Kit enables users to develop powerful, fully customized C++ solutions for numerous machine vision applications. With CVL, users can modify the sequence of vision processing operations, define custom tools, and link vision tools together. The Developer's Kit also enables users to perform a variety of specialized vision functions such as pixel-level processing and complex image buffering.
VisionPro® is a suite of machine vision software tools that support both the Microsoft Visual Studio® .NET and COM/ActiveX-based programming environments. VisionPro users can easily integrate third-party objects into their vision applications and readily integrate vision applications with their automation system. VisionPro offers the power and flexibility of advanced programming, and the ease of development provided by graphical programming environments such as Visual Basic and C#. This enables OEMs and system integrators to quickly develop and deploy powerful machine vision applications.
In addition, Cognex offers a wide-range of vision tools for object location, defect inspection, gauging, and part identification, as well as a variety of specialized functions.
[Bruce] By incorporating the processor on the frame grabber, developers can offload the per pixel processing from the CPU to the Frame Grabber using the Software SDK. This allows PCI Bus interface (because the data rates are greatly reduced).
4. What are some frame grabber features that are of particular importance for machine vision applications and why?
[Andrew] The DVR Express frame grabbers record video data locally while sending a live video stream over the computer PCI bus for real-time display and processing by the host computer. In machine vision applications, this live video stream can be used for process monitoring or real-time process control. Trigger inputs and outputs on the DVR Express frame grabbers provide additional control interfaces in the machine vision system.
[Stephane] Reliability - Intelligent Frame buffer (tap reordering, buffer for transfer) no data lost.
Flexibility - PCIe 1x 4x and PCI 32-64bit, acquisition modes, independent dual base, medium, I/Os...
Performance - 85MHz CL drivers, PCIe 4x, HW 5x5 non-linear bayer interpolation.
- Very low trigger to acquisition latency
- Camera variety allowing customers to use their camera of choice.
- A large frame buffer (FIFO) to pipeline images in high speed applications.
- Onboard DMA channels for fast image transfer to the PC
- Onboard I/O - Including triggers, strobes, encoder inputs for line scan applications, and general-purpose I/O to reduce system cost.
[Bruce] Multichannel applications, full programmability, advanced processing possible -- direct IDE interface - for multi-card recording and playback. For test, inspection and documentation applications, the automated Video Content Analysis and SDS data recording features allow large scale systems to be deployed easily.
[Philip] What differentiates machine vision frame grabbers from other commercially available frame grabbers are the speed and reliability of the acquisition process. Machine vision frame grabbers are designed specifically to acquire images at rates from 100 Mbytes/sec to 1 Gigabyte/sec, and to do so reliably (without losing image data). Additionally machine vision frame grabbers include machine vision I/O capability such as trigger inputs, strobe outputs, actuator outputs, and camera control signals.
5. What are the specific application issues that one must be attentive to when applying a frame grabber? For example, camera, input/output, resolution issues, etc.?
[Stephane] For area scan cameras the customer has to select just the correct camera model, predefined from Leutron, and the acquisition mode (continuous live, reset/restart on external trigger, flash output, shutter etc. For line-scan there exists extensive real time support to synchronize the application via encoder and digital I/Os with the image acquisition. Line scan applications can be mixed with area scan camera on a dual input frame grabber.
[Bashar] Typical machine vision system flow is Trigger -> Acquisition -> Image transfer -> Image processing -> Result communicated over I/O. The following considerations must be taken into account in each phase:
Trigger phase: Frame grabbers that don't offer onboard trigger support limit the usability and the applications that the frame grabber is able to solve. In addition, minimal delay is required from the time a trigger is received until the time an image is acquired. Moreover, frame grabbers must be able to detect and report missed triggers.
Acquisition phase: The user must determine the type of camera needed for the application (area scan or line scan). Then determine the resolution of the camera needed for the application. This depends on the area the customer plans to inspect. The higher the resolution of the image, the more robust the vision tools will run but the longer it takes for them to run. The customer must also determine the frame rate of the acquisition. Applications vary from 1 frame per second or lower to hundreds of frames per second.
Image transfer phase: The user must insure that the frame grabber utilizes DMA technology in order to spend the least amount of time transferring images to the PC. Moreover, the PC must be able to keep up with the size of the data being transferred over the PC's bus.
Image processing: The user must select vision tools and apply them properly. Vision tools are not the same - this is the area that customers must pay very special attention to. Cognex has the best pattern-matching tool on the market and a comprehensive set of supporting vision tools.
Result communication: Onboard I/O tremendously reduces the cost of the overall system because the user doesn't have to purchase third party I/O and integrate it into the vision system. The availability of Optically Isolated I/O greatly reduces the effects of noise on the I/O signal and the chance of signal corruption.
[Bruce] Camera inputs are NTSC/PAL, interface is PCI (Windows PCI SDK driver provided).
[Philip] One must pay particular attention to the type of camera being utilized in the application, to ensure that the output characteristics do not exceed the input capabilities of the frame grabber. Therefore, issues such as resolution, bit depth (including whether it is a color or monochrome camera), and pixel clock speed are important variables to consider. Attention should also be focussed on any special camera modes that are being considered (special triggering modes, binning, long or extremely short exposures) to ensure that they are compatible with the frame grabber being evaluated.
[Andrew] The video data rate is a fundamental consideration in choosing the right frame grabber. The video data rate is calculated as (Horizontal Frame Resolution x Vertical Frame Resolution x Bytes per Pixel x Frames per Second). If a conventional frame grabber is to be used, it must be capable of sustained data transfer at the video data rate over the computer peripheral data bus. If the application requires real-time image processing, the application software must be capable of processing video frames at the frame rate of the camera. If the application requires real-time storage, our DVR Express frame grabbers should be considered as the frame grabber in the system.
6. What is it that you require from a prospective buyer of a frame grabber for an industrial application to assure that it will satisfy their application?
[Bashar] We ask customers for plenty of samples so that a Cognex Application Engineer may perform a comprehensive free of charge evaluation on the customer's parts to determine whether machine vision is the appropriate solution for the application. We provide customers with a completed evaluation report and make a recommendation on which frame grabber, camera, lens and vision tools they should use.
[Bruce] Summary system specification showing the inputs/outputs and description of the processing they wish to port to the Grouper Intelligent Frame Grabber.
[Philip] DALSA requires a description of the application, the camera being considered for the application and any special modes of the camera that are required.
[Andrew] Knowledge of the camera model is the first step in determining whether or not a frame grabber is suitable for an application. In addition the special camera settings and modes of operation in the application must be known. If a conventional frame grabber is to be used, the interface to the host computer system, PCI, PCI-X or PCI Express, must be considered to ensure it can sustain the peak video data rate.
[Stephane] We need to ensure camera interface by getting camera specs (if we don't know the camera already), we need the OS/processor for driver compatibility. As an option it is always nice to understand the acquisition mode, especially if it is a complicated one.
7. What are the skills required to integrate a frame grabber?
[Bruce] Host (PC) application development and familiarity with PCI SDK integration. For Porting -- C code and potentially some C-like assembly (easier than MMX).
[Philip] It all depends on the complexity of the application. Getting a frame grabber from DALSA up and running with any camera is relatively easy and requires basic computer knowledge, and, therefore, no special skill sets. Therefore, for basic grab and display applications, most computer users would get the frame grabber and camera functional in about ten minutes.
However, if the application is complex, with a high degree of interaction between different system components, it is a different story. Therefore, integrating a frame grabber into the overall system which may include special lighting, lens control, camera control, motion controllers, and actuators does require system engineering level skill sets, and the ability to program special functions of the frame grabber.
[Andrew] Since the frame grabber is integrated as part of a system, the integrator must have a good working knowledge of the overall system. Most importantly, a thorough knowledge of PC system architecture and its peripheral interfaces is needed. A familiarity with electronics is helpful in understanding the interface between the frame grabber, the camera and potentially other devices external to the computer. Experience with machine vision software is useful. If customized image processing procedures must be developed, then an understanding of image processing theory is essential.
[Stephane] Depends on the application, from computer notions to software programmer. The minimum is a basic understanding of the camera and acquisition mode and the ability to program VB using active-X tools.
[Bashar] Depending on the application complexity, Cognex’s development environments range in the level of skill required. VisionPro’s Quickbuild development environment requires no programming skills to develop an application. Through a series of mouse clicks, a user can setup Quickbuild to acquire, display, and process and image with vision tools. As the application’s requirements get more complex, users can use Microsoft’s Visual Basic to build their applications in VisionPro or use C++ to build their applications in CVL.
8. How do you support your frame grabber product line? E.g. help with set-up? Post service support? Training? Warranty? Documentation?
[Philip] As a testament to our level of support, DALSA implemented the OEM Partnership Program that features an extensive list of support services from basic on-line or phone support, to on-site visits and training from our application engineers. Additionally we offer customization services which ranges from minor tweaks to frame grabber functionality to custom FPGA processing routines.
Also of concern to OEMs is long-term supply of components, emergency inventory supply, and version control. These are concerns are all addressed by our OEM Partnership Program.
[Andrew] To support our customers in system integration, we will pre-install the frame grabber in the computer system and provide instructions for connecting the system to the camera and potentially other external devices. Free technical support by telephone and email is provided to assist customers with setting up our hardware. Product manuals guide the user through the operational details of the hardware and software.
[Stephane] Most of the support is done by documentation online, Email and phone calls.
[Bashar] Cognex provides pre-sales evaluations, post-sales phone and web support, warranty on hardware, on-site or off-site training and comprehensive documentation with plenty of examples. Moreover, Cognex has a dedicated Vision Solutions group that is willing to travel to a customer’s site to help them with their most demanding applications.
[Bruce] Cradle provides tools, tools support and training. 3rd party developers are available for porting. Reference Design Kit is available to allow OEMs to manufacture the boards themselves.
9. Are there some emerging technology changes associated with the underlying technology used in frame grabbers that will impact the performance of frame grabbers? What will those impacts be?
[Andrew] New fiber channel controller chips were introduced earlier this year that operate at speeds up to 4 Gbps. The increased bandwidth of the hard disk connection will enable us to introduce new frame grabber products that can record video at higher data rates. This opens the door for the recording of high-speed cameras that operate at ‘‘1K Cubed’‘, that is an image size of 1024 x 1024 pixels and a frame rate of 1000 frames per second.
[Bashar] PC bus technology has a tremendous impact on frame grabbers. As you probably know, ISA bus technology is almost phased out and replaced with PCI. Most recently, PCI-Express was introduced to workstation class PCs and is expected to appear on personal PCs this year. The bigger the ‘‘pipe’‘ is or as PC's image transfer capability increases, the more images or the higher resolution the frame grabbers can send. Cognex will continue to look for ways to design and develop frame grabbers that can transfer images at the fastest rates possible.
[Bruce] Intelligent Frame Grabbers, all real-time Video Content Analysis and measurements, as well as recording the results. For Test and Inspection applications, this will allow more channel density and lower cost per channel than using larger PCs and faster busses. 16 NTSC Channels raw would be over 320 MBytes/second. By providing the Video Content Analysis in real-time on the frame grabber, the output data rate can be reduced to below 1 MByte/sec - so a single PC can support multiple frame grabbers.
[Philip] The biggest emerging technology (in fact it is pretty much here) is the PCi-Express, or PCIe bus. This serial digital protocol promises to increase the performance of PC systems, and yet continue the promise of improving prices.
Another underlying technology that promises to change the way things are done is the use of CMOS sensors for machine vision applications. The high frame rates and random access are two features that are just now being exploited.
[Stephane] The cost of a technology limits its use in a frame grabber. So the realization at an acceptable cost reduces the performance. The key areas are:
- Host computer interface
- Onboard memory/buffer
- Front end performance (digitalization and interfacing)
- and Onboard preprocessing
10. What are major trends you see associated with future frame grabbers?
[Bashar] You'll see more standardization on connectors and cables. Frame grabbers will continue to become more efficient and more cost-effective.
[Stephane] Two markets:
- Multiple cameras: Frame grabbers ensure that data from many cameras are transferred safely into host with very low and predictable latency: |
‘‘latency’‘ in these cases is understood the time from pixels or frames are arriving on the frame grabber until they arrive in the host memory. With PCI grabbers and proprietary software drivers this time is very
There are currently no comments for this article.
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.