Building a new system architecture for intelligent surveillance video with FPGA
As high-definition (HD) security monitoring and video processing systems continue to evolve, the demand for powerful and efficient processing devices has increased significantly. Traditional single-chip DSP solutions are no longer sufficient to handle full-frame-rate video processing. While multi-chip or CPU+DSP configurations can meet some requirements, they often come with challenges such as higher PCB costs, resource inefficiency, and system instability. In response, a new system architecture that leverages the flexibility of FPGAs to load video processing engines is gaining attention in the industry.
Cost and Performance Challenges
Jiang Yungui, a senior system architect at Xilinx Asia, noted that the market for intelligent video analytics in surveillance is expected to reach $800 million by 2012. For designers of video surveillance equipment like cameras and DVRs, single-chip solutions face limitations in performance, cost, and power consumption. By using Xilinx’s Spartan-3A DSP FPGAs, which offer programmability and cost-effectiveness, companies can implement multiple monitoring functions—such as object tracking, virtual barriers, and demographic analysis—on a single chip, making it a more viable solution for modern surveillance needs.
To support this, FPGAs require robust video processing engines. Eutecus, Inc., based in Berkeley, California, developed Cellular Vision Technology (CVT), a next-generation image stream capture and processing solution that integrates parallel architectures and advanced algorithms. Stephen D. Hester, CEO of Eutecus, explained that CVT was inspired by the human visual system, enabling them to provide image processing IP for FPGAs and other devices in the security and video surveillance markets. “We faced challenges in developing our Multicore Video Analysis Engine (MVE),†he said. “Our first generation used TI’s DaVinci SoC, but for the second generation, we needed more processing power and better system integration. Using multiple DSP chips wasn’t cost-effective, so we looked for a single-chip solution that could carry over previous features while adding new capabilities.â€
Spartan-3A Based Processing Architecture
After thorough evaluation, Eutecus selected the Xilinx Spartan-3A DSP3400A. This device contains 126 XtremeDSP DSP48A slices, delivering up to 30 GMAC of DSP performance, making it ideal for video analytics applications that require both high performance and low cost. The MVE can process HD video at full frame rates while consuming less than 1W of power. It offers an easy-to-configure, compact design that combines a built-in parallel multi-core architecture with embedded video analysis algorithms, simplifying hardware development and programming. This allows the MVE to perform simultaneous event detection and parallel classification algorithms efficiently.
Portability across platforms is a key consideration in system design. Xilinx’s EDK Embedded Development Kit supports a dual-processor hardware architecture based on the MicroBlaze embedded processor, similar to TI’s DaVinci platform. This makes it easier for Eutecus to migrate their designs between platforms without significant rework.
“By embedding MVE in the Xilinx Spartan-3A DSP FPGA, we can deliver complex video analysis capabilities to smart cameras or network devices like DVRs, all in a single chip,†Jiang Yungui explained. “FPGAs allow us to add more video analysis features that meet customers’ evolving needs. With the same hardware platform, we can create various derivative products. Eutecus uses VHDL to develop multiple analysis accelerator engines and integrate these specialized cores into the C-MVA coprocessor. This approach enables engineers to reuse dual MicroBlaze systems, generating different FPGA programming files and creating a scalable solution that adapts easily to a wide range of video analytics applications.â€
RFID System,RFID readers,RFID Readers with IO-Link
Kunshan SVL Electric Co.,Ltd , https://www.svlelectric.com