Vision Systems

Increasing Image Acquisition Rates for Bandwidth-Hungry Applications

“Bandwidth-hungry” vision applications run into a bottleneck transferring image data. But now PCI Express, Camera Link and FPGA technologies have been implemented to increase image acquisition and processing rates.


  • Page 1 of 1
    Bookmark and Share

Electronics component manufacturers must continually improve productivity and quality to remain competitive. Motion control and machine vision play an important role in this effort by providing automated inspection capabilities that are more reliable and scalable than legacy methods. As manufacturers require increased image throughput and more sophisticated image processing, transferring the data from the camera to the PC, or bandwidth, often becomes a limiting factor that must be addressed.

Standard VGA resolutions and 30 frame/second capture rates were sufficient for most production lines in the past. However, industry requirements are now demanding an increase in the dimensions scanned, e.g. line scan, 3D inspection, OCR, barcode and 3D barcode. In addition, there is a push for an increase in the production line conveyer speed, and more complex image processing. Meanwhile, bandwidth has remained the main bottleneck for PC-based machine vision systems.

Camera Link, a new camera standard, was introduced to address the bandwidth issue. Combined with the PCI Express bus, Camera Link image solutions provide unprecedented data transmission for machine vision applications. Figure 1 shows the primary components of a typical computer-based machine vision system. These components include a camera with lighting, cabling, plug-in frame grabber card and computer. Camera Link was developed as an open specification for cameras and frame grabber cards to be used in such machine vision systems. The Camera Link interface specification was defined by the Automate Imaging Association, an industry group consisting of industrial camera, cable and frame grabber card manufacturers. Camera Link utilizes Low Voltage Differential Signaling (LVDS) technology to transmit digital data. Both a parallel-to-serial transmitter and a serial-to-parallel receiver are used to transmit image data.

Camera Link

The Camera Link standard provides for three configurations: base, medium and full signaling modes. Image data from the camera is transmitted through a designated port to the corresponding port connected to the frame grabber (Figure 2). The base configuration consists of a set of transmitter/receiver pairs (24 bits of video data organized into three 8-bit ports: A, B, and C. Additional signals include frame valid (FVAL), line valid (LVAL), data valid (DVAL) and spare signals plus one clock signal, one serial communication channel and four LVDS general-purpose camera control signals.

The camera control signals are typically used with external triggering to capture images in real time. The clock runs at rates between 20 to 85 MHz for an image data transmission rate of up to 2.04 Gbits/s. The medium configuration adds a second set of transmitter/receiver pairs to double the number of image data bits to 48 bits, which also doubles the number of 8-bit ports to six: A, B, C, D, E and F. The image data transmission rate is also doubled to 4.08 Gbits/s. The full configuration further increases the number of image data bits to 64 bits for eight 8-bit ports: A, B, C, D, E, F, G, and H, and improves image data transmission rates to 5.44 Gbits/s. The Camera Link data routing is illustrated in Figure 3.

Of the currently available camera interfaces (such as analog connections, USB, FireWire, GigE), Camera Link provides the following advantages:

• Highest bandwidth: Camera Link streams image raw data at rates up to 5.44 Gbits/s over dedicated point-to-point link topology without protocol overhead.

• Noise reduction: The minimal LVDS signal of the Camera Link specification increases image throughput between the frame grabber, cable and camera.

• Real-time signals: There is no network latency or protocol overhead with Camera Link.

• Reduced CPU workload: The Camera Link specification includes standard chipsets that can be used for on-camera image processing. Camera Link frame grabbers also utilize direct memory access (DMA) for optimal data transfer efficiency. Using DMA, the transmission of image data from the frame grabber to the memory of the host PC is executed without utilizing any CPU resources of that host system.

• Support for line scan cameras: Camera Link offers the camera control, serial communication and data streaming methods needed for line scan applications. Most line scan cameras available today already support Camera Link

PCI Express

Computer systems used in machine vision applications may now utilize the new PCI Express (PCIe) bus technology. The PCIe bus implements a serial packet-based protocol along with a switch-based topology to provide high bandwidth and robust point-to-point interconnects, which is a significant improvement over the predecessor PCI bus that offered bus-based connections where all the devices shared the same 32- or 64-bit parallel bus.

The PCI Express bus maintains complete software compatibility with the existing base of operating systems, PCI drivers and software, in addition to offering a scalable bus bandwidth. The physical layer of a PCIe link is a set as serial lanes in groups of 1, 2, 4, 8, 12, and 16. Each lane provides a 2 Gbit/s data transmission rate, which equates to a rate of 8 Gbits/s for a 4-lane link. The PCIe lanes also provide dedicated bus connections whereas bandwidth on the PCI bus is shared among all PCI devices and some system functions. In other words, the PCIe bus provides a dedicated link for image data transmission so that bandwidth is not shared during transmission.

FPGA-Based Image Processing

As the amount of transferred image data increases, additional computing power is required in computer-based machine vision systems to maintain performance. As mentioned above, the majority of frame grabber cards support DMA to reduce CPU loading. Image processing on the host computer also loads the CPU, and can often lead to a system bottleneck.

In the past, the only solution for image processing bandwidth was to place processing in the form of an ASIC onto the camera or frame grabber. However, ASIC solutions significantly increased development costs, delayed time-to-market and limited processing flexibility. More recently, users have taken advantage of computer processor advances, such as multicore, to increase image processing bandwidth. But processor solutions also have limitations in that they are unable to integrate other hardware I/O signals.

The use of a programmable component in the form of an FPGA is able to address the limitations of both ASIC and processor-based throughput solutions. An FPGA provides a set of logic elements through which the pre-processing functions are executed on the frame grabber card, in parallel with the image acquisition and without causing any loading on the host CPU. The combination of the CPU and FPGA provides the proper balance between the performance and the cost of the machine vision system. Advantages of incorporating an FPGA include:

• Flexibility: The FPGA is a processing core that provides programmable and configurable functions.

• Parallel processing: To increase performance, an FPGA provides parallel computing power for image pre-processing algorithms (such as generic matrix multiplication) to speed up calculation.

• Reusability: An FPGA can quickly integrate intellectual property (IP) blocks without any modification of the logic program, thus reducing time-to-market.

• Memory Access: Certain image pre-processing tasks need to perform operations between multiple frames or lines. Reference image or lines are stored in the system’s memory where the FPGA is able to access these memories in parallel.

• Implementation of PCI Express: Many FPGA vendors offer PCI Express IP cores for the implementation of the PCI Express protocol. Utilizing such FPGAs will reduce the hardware cost of the frame grabber.

FPGAs are able to perform pre-processing functions such as shading correction, color space conversion, image rotation and look-up table operation. FPGAs can also provide enhanced images on-the-fly, ready for further processing on the host CPU, further supporting the concept of a proper balance between system performance and cost.

The combination of Camera Link, FPGA and PCIe technologies can provide the high bandwidth and high transfer speed required by advanced machine vision applications such as line scan and 3D inspection.

For example, flat panel surface inspections typically require the simultaneous capture of multiple image channels. A PCI Express-based system supports multiple frame grabber boards and provides each board with a dedicated link to the host memory. Because the bandwidth of the PCI Express bus is guaranteed, each frame grabber can operate at full speed without sharing bandwidth. Thus the frame grabbers will not affect the performance of the other frame grabbers in the system, resulting in an inspection system with maximum throughput. Yet, as the image quality and image processing speed in machine vision applications increase, performance requirements may not be met solely through the host PC; the addition of an FPGA offers both increased performance and design flexibility.

Camera Link offers the optimal solution for bandwidth, real-time signaling and robust image transmission of the currently available machine vision solutions. Overall performance can be further improved when this technology is combined with other technologies, such as FPGA, to provide the system developers with the high-speed and high-resolution capabilities they need at a price they can afford.

ADLINK Technology
Irvine, CA.
(949) 727-2099.