string(27) "" Next Page

Grabbing The Frame | Silicon Software

  • By Jason Stockwell

Almost written-off by the introduction of new technology, the venerable frame grabber is alive and well. Here we look at some of the major players.

Silicon Software

An update from Silicon Software.

Multi-tasking Frame Grabbers for High-speed Applications

“More powerful sensors with ever higher resolutions and increasing demands on the entire image processing system as a whole require more bandwidth resp. data throughput for the camera interface. Among the high-speed interfaces, CoaXPress is one of the most important, and its predicted market share will continue to increase in the future. The current CoaXPress frame grabbers of Silicon Software’s “microEnable marathon” product line support CoaXPress version 1.1.1. The next generation will be designed for CoaXPress version 2.0 with a configuration up to CXP-12.

“A special high-speed frame grabber of the “CNN ready” series with Camera Link interface is already available for deep learning applications with convolutional neural networks (CNN). Keeping pace with the expansion of existing and the availability of completely new interfaces, new frame grabbers for the camera interfaces of the next generation will be released: CXP-12, 10 GigE and NBASE-T. Besides “power over” feature, all frame grabbers unify image and signal processing, image pre-processing and camera control via the on-board FPGA processor.”

CoaXPress Frame Grabbers

“One, two or four compatible CoaXPress camera types can be connected respectively to the image acquisition and processing boards. The frame grabbers support colour (RGB and Bayer) and monochrome area, line scan and CIS cameras and up to 25 GB/s incoming bandwidth They support the CoaXPress configurations CXP-1 to CXP-6 and are well suited for high performance cameras, industrial multi-device and multi-camera solutions.

“The variant with four ports can be configured from one high-speed camera with four inputs to four different CoaXPress cameras at the same time with a multitude of pixel formats and bit depths. The V-series is graphically programmable with the software VisualApplets in short time using data flow models to realize specific image processing applications with real-time, deterministic and low-latency behaviour. Existing FGPA hardware code (created with VHDL or Verilog) can also be integrated using VisualApplets Expert.

“All CoaXPress frame grabbers include FPGA-based image pre-processing (e.g. Bayer filter, lookup tables and white balance) at a very high frame rate, minimizing CPU load and accelerating the computer’s overall system performance. This guarantees a cost-efficient system setup with increased application performance. Based on PCI Express x4 (Gen 2) they take advantage of the Silicon Software DMA1800 technology, combining the maximum data throughput of 2.5 GB/s at 4-channel operation with intelligent data reduction by image pre- or post-processing.

“General Purpose Input/Output signals can be used via two GPIO connections that are independently configurable (slot bracket GPIO and onboard GPIO with an interface to TTL or Opto trigger boards). With these trigger signals it is possible to internally synchronize several frame grabbers or peripheral devices in series (daisy chain), amongst other techniques.”

Special FPGA Based Deep Learning Board

“The programmable “CNN ready” Camera Link frame grabber equipped with a runtime license for deep learning under VisualApplets fulfils the requirements for high computing performance and bandwidth and offers a more powerful FPGA. The board calculates larger neural nets with bandwidths over 200 MB/s without delay.

“The integrated FPGA features high parallelism of processing, low thermal power, deterministic latencies and long market availability. The frame grabber offers a more cost-effective, energy efficient and faster solution than an industrial GPU comparable solution.

“Using graphical FPGA programming with VisualApplets, suitable net architectures can be integrated and pretrained configuration parameters for the weights of the networks can be imported.”

Get more stories like these Subscribe Sign in