A Project-Based Learning Approach for Teaching Edge AI and Embedded Systems: An Interdisciplinary Case Study on Vision-Only UAV Detection Using YOLOv8 on Raspberry Pi 5

Abstract

Engineering education in artificial intelligence, embedded computing, and computer vision often produces students with strong theoretical foundations but limited experience deploying real systems. This paper presents a twelveweek, project-based learning exercise in which students built a vision-only UAV detection system on a Raspberry Pi 5. The pipeline incorporates CLAHE and gamma correction preprocessing, a pre-trained YOLOv8 flying object detector, monocular distance estimation via a pinhole camera model, pantilt servo tracking, infrared illumination using a NoIR camera, MJPEG streaming via Flask, and Telegram alerts — all running on consumer-grade hardware. Students worked in teams across five modules, each responsible for one subsystem while remaining accountable for full system integration. Hardware testing achieved 86% detection precision at 4.2 FPS, with MJPEG latency of approximately 180 ms, pan-tilt settling time under 0.8 seconds, and Telegram alert delay under 2 seconds. Daytime performance consistently exceeded nighttime results despite IR illumination. These outcomes demonstrate that multi-subsystem edge AI deployment is feasible on constrained hardware. Postproject feedback indicated that the most durable learning gains came not from mastering individual technologies, but from reasoning through how changes in one subsystem propagated across the entire architecture.