Speaker
Description
Object detection is a fundamental capability in modern computer vision systems, supporting applications such as autonomous vehicles, robotics, smart surveillance, and edge AI. However, deploying deep learning–based detection models on resource-constrained platforms remains challenging due to high computational demands and energy consumption. Field-programmable gate arrays (FPGAs) provide a promising solution by enabling low-power, real-time acceleration of vision workloads at the edge. This project explores the feasibility of implementing and optimizing object detection pipelines using the PYNQ FPGA platform, with an emphasis on hands-on semiconductor workforce training and practical edge AI deployment. Inspired by recent demonstrations of real-time edge detection on PYNQ, we extend FPGA-based image processing toward more advanced detection tasks, including lightweight convolutional neural network (CNN) inference and hardware-accelerated preprocessing. The study investigates how FPGA programmable logic can support key components of object detection, such as feature extraction, edge-enhanced filtering, and model acceleration. The project also evaluates trade-offs between software-based Python execution on embedded ARM processors and hardware-based acceleration using FPGA overlays. The long-term goal is to develop an accessible framework for integrating computer vision with FPGA-based acceleration while contributing toward efficient, real-time object detection at the edge. This work supports the mission of semiconductor and AI workforce development.
| Academic or Professional Status | Undergraduate Student |
|---|