Date of Award

6-4-2026

Document Type

Thesis

Publisher

Santa Clara : Santa Clara University, 2026

Departments

Computer Engineering; Computer Science and Engineering

First Advisor

Sean Choi

Second Advisor

Michael Schimpf

Third Advisor

Radhika Grover

Abstract

This thesis presents the development of a low-cost embedded image processing system for autonomous vehicle perception using the PYNQ-Z2 FPGA board. The project explores whether compact FPGA-based systems can provide efficient, low-latency perception for small-scale autonomous applications such as delivery robots, warehouse robots, and drones, where power efficiency and affordability are important design constraints.

The system uses Python and OpenCV for image preprocessing and color-based traffic sign segmentation, while a modified LeNet-5 convolutional neural network performs traffic sign classification. The model was optimized and deployed onto the FPGA using Vivado HLS and Vitis to accelerate inference on resource-constrained hardware. The classified outputs were integrated with motor control logic to demonstrate autonomous sign reaction behavior, including real-time stop-and-go responses.

Experimental results were evaluated based on classification accuracy, inference latency, and overall system responsiveness across software and hardware implementations. While challenges related to FPGA integration and memory communication were encountered, the project demonstrates the feasibility of using low-cost FPGA platforms for realtime embedded computer vision in autonomous systems.

Download

Share

COinS