Efficiency and Performance Tradeoffs in FPGA-based Embedded Computer Vision Applications

This research study delves into the efficiency of artificial intelligence (AI) algorithms on embedded hardware, focusing on field-programmable gate arrays (FPGAs). Amidst the growing environmental concerns due to AI's high energy demands, this research explores using FPGAs as a more sustainable alternative to traditional processing units like GPUs and CPUs. The study is framed around three research questions: Firstly, it investigates the energy consumption patterns of AI algorithms in embedded hardware. Secondly, it examines strategies for optimizing AI model sizes while maintaining performance. Finally, the studies explore various hardware and algorithmic optimizations to further reduce AI models' energy footprints. Initial results with tools like HLS4ML and Vitis-AI, which facilitate the integration of AI models into FPGA platforms are presented. This research aims to provide insights into the trade-offs between energy efficiency and performance in embedded AI applications, presenting initial findings on the usage of advanced AI models on FPGA hardware.