每秒42帧 - Photonfocus最新高光谱千兆以太网视频摄像头让你看到不曾看过的精彩！

每秒42帧的视频流 —— Photonfocus最新高光谱千兆以太网视频摄像头让你看到不曾看过的精彩！ Xilinx 旗舰产品 Spartan-6 为视频处理及定制化客户需求助力！

The just-announced hyperspectral Photonfocus MV1-D2048x1088-HS02-96-G2 GigE video camera produces a 42-frames/sec video stream based on 25 spectral pass bands from 600nm to 975nm (from orange/yellow through near infrared), resulting in a 10-bit grayscale video representation. This hyperspectral GigE video camera is based on an IMEC snapshot mosaic CMV2K-SSM5x5-NIR sensor, which in turn is based on the CMOSIS CMV2000 2048x1088-pixel CMOS HD image sensor.

IMEC produces these sensors through wafer-level processing that deposits pixel-level Fabry-Perot interferometer filters in a 5x5-pixel mosaic onto the prefabricated CMOSIS HD image sensor. The 5x5 filter mosaic produces the 25 spectral pass bands. Here’s an illustration of the resulting HD video imaging sensor:

IMEC Hyperspectral HD Video Imaging Sensor

And here’s the resulting spectral sensitivity curve for the Photonfocus MV1-D2048x1088-HS02-96-G2 GigE video camera based on this IMEC sensor:

Photonfocus MV1-D2048x1088-HS02-96-G2 Hyperspectral Sensitivity Curve

Photonfocus simultaneously introduced a second hyperspectral camera, the MV1-D2048x1088-HS01-96-G2-10, equipped with the IMEC CMV2K-LS100-NIR line-scan wedge sensor. Photonfocus integrates these IMEC hyperspectral imaging sensors into its GigE cameras and provides an API-level interface to a Gigabit Ethernet imaging network that can be used by a variety of imaging software tools including MVTec’s Halcon and National Instruments’ LabVIEW. The two Photonfocus hyperspectral cameras employ a Xilinx Spartan-6 LX75 FPGA for internal video processing and the company can implement additional imaging features in the camera’s FPGA for custom applications at customer request.

Color video cameras split light into three bands—usually red, green, and blue—using a Bayer mosaic filter to produce an interpolated video image that resembles what’s seen by the unaided human eye. Hyperspectral imaging cameras split incoming light from a scene into both visible and invisible spectral bands (many bands in the case of hyperspectral cameras) so that the camera can see what the human eye cannot. Hyperspectral cameras find use in diverse imaging applications including:

• Remote sensing for agriculture and environmental monitoring
• Food processing
• Mineral extraction and processing
• Medical diagnostics
• Surveillance and military sensing
• Astronomy

Coincidentally, Xcell Daily has covered Photonfocus video cameras before, but they were operating unsung in the background (see “How would you like to get a 10x speedup in your image-processing or optical-inspection system using the Zynq SoC?”). The video that appears in this earlier Xcell Daily blog post shows Dr. Klaus-Henning Noffz, CEO and founder of Silicon Software, discussing a high-speed, optical-inspection application at last November’s SPS Drives show in Nuremberg, Germany. This application was implemented with Silicon Software’s Visual Applets development platform. As the video shows, Visual Applets was able to boost the application’s image-recognition performance from 8 to 92 frames/sec by moving some of the Visual Applets video-processing algorithms from software running on a fast microprocessor into the programmable logic hardware in a Xilinx Zynq SoC running the application.

Here’s the video once again in case you missed it the first time:

Of course, it’s not sufficient to coax a video-processing algorithm to run that fast if the frame rate from the camera doesn’t support the faster speed. Those two cameras you see above the spinning drums to the right of Dr. Noffz in the demo video are Photonfocus DR1-D1312-200-G2 high-speed, double-data-rate video cameras capable of delivering 1312x1082-pixel images at 135 frames/sec over a GigE interface. That’s almost 2x the frame rate of a conventional GigE video camera. How does this high-speed Photonfocus GigE camera achieve such a fast video performance? In part, from the high-speed video processing being performed in the camera’s programmable logic, contained in a Xilinx All Programmable device.