FPGA-Based Hardware/Software Codesign for Video Encoder on IoT Edge Platforms

Abstract

Recently, image/video-based applications have been widely used for many domains, such as traffic, medical, or robotics. In this context, IoT-based systems with video processing implemented on edge devices have been applied extensively, for example, surveillance, object detection and monitoring, or intelligent home/building systems. Although algorithms for images/videos encoder with various standards produce good results, they usually require high computing power that is limited with edge platforms. Therefore, we propose an efficient hardware/software codesign approach to accelerate the video encoder process with FPGA-based IoT edge computing platforms in this work. The design flow exploits the effectiveness of the high-level synthesis process to quickly and efficiently generate FPGA-based hardware accelerator cores to improve system performance. We use the H.264 encoder as our case study to verify the proposed design flow and evaluate the accelerator ability compared to general-purpose processors. An FPGA Ultra96-v2 edge computing board is used for conducting experiments with a certified dataset. Experimental results show that we obtain speed-ups by up to 14.9$\times$ compared to a 4-core ARM processor. Our experimental system also saves up to 6.24$\times$ energy consumption compared to the ARM processor.