At the recent Embedded Visual Conference, advanced FPGA / DSP design engineers from Xilinx Partner An Fuye (Avnet) Mario Berger showed a dual camera to collect images and achieved fused demonstration Demo. The hardware platform uses the AVNET Picozed Som (on-chip system), the core integration is the Xilinx Zynq Z-7030 SOC, and also uses a Module of the FMC extension board of the Picozed embedded development kit. The two cameras used are the Python-1300-C color image sensor of the FMC interface. The resolution is 1280x1024, the configuration is flexible, high sensitivity, high performance, mainly used in industrial image acquisition applications, and Avnet) company design produced. Another camera uses a FLIR infrared thermal imaging sensor, an infrared video stream of 60x80 resolution, and implements communication and data transmission with Picozed SOM through the PMOD interface.
Figure 1: Avnet Dual-camera image fusion application module design
First, the color image is subjected to Sobel Edge Detection Processing, and then the infrared image is stretched and the size is re-adjusted so that the two images can be fused. During the development process, the algorithm of the Xilinx Revision Stack resource, Sobel edge detection and re-adjusting image size is directly from the algorithm library provided by Revision Stack, and the image fusion part is an engineer Mario Berger using a C language because of these three algorithm modules. They are all calculated intensive tasks, so use Vivado HLS to create a hardware accelerator module, and then use the SDSoC tool to implement the communication connection of the accelerator module and the ARM processor to generate a software interface driver.
What is the performance of the entire system? We make a comparison with the Sobel edge detection module. If the non-hardware accelerated Sobel algorithm, the load instant of Zynq Z-7030 SOC's ARM Cortex-A9 processor is 100%, and the output is only 1 frame / second. If the hardware accelerates, the image output can reach 30 frames / sec, and the load of the ARM processor is also reduced by 80%, and the specific test is shown in the following video:
The Xilinx Revision Stack includes rich platforms, algorithms and application development resources, supporting the most popular neural networks, developing, providing a comprehensive industrial grade framework, including Caffe facing machine learning, and OpenVX facing computer vision. Xilinx Revision Stack supports more widely unloating embedded software and system engineers that are very hardware design expertise, making it fused with machine learning, computer vision, sensor, so that visual orientation can be developed more quickly.
Original link: https://www.eeboard.com/news/xi/
Search for "Love Bo.com" to pay attention, daily update development board, intelligent hardware, open source hardware, activities, etc., you can make you master. Recommended attention!
[WeChat scanning picture can be directly paid]
Technology early know:
Black Technology: Manchester explosion about 20 people dieted black technology to help counter-terrorism equipment upgrade
Microsoft releases new Surface Pro, foreign media: product strategy is as follows apple
Google Tensorflow person in charge: Baidu open source is beneficial to Chinese developers
Sharing bicycle or leading the narrow band network broke out three operators
ADI launches small isolate gate driver to provide next-generation power switch technology solutions
Our other product:
