<![CDATA[Waste management in Indonesia is still characterized by a high volume of improperly managed waste and low source-level segregation, causing recyclable materials to mix with other waste streams and reducing their recovery value. This situation calls for a sorting system that is effective, fast, and affordable, while also providing real-time operational information to support on site decision-making. This study presents an integrated computer vision approach using YOLOv8 deployed on a Raspberry Pi 5 with a Camera Module 3, connected to a real-time information system via Server-Sent Events (SSE) for monitoring and analytics. The methodology includes constructing a labeled dataset in YOLO TXT format, training a YOLOv8n model, deploying edge inference, and developing a backend API to receive detection outputs and stream them to a dashboard in real time. The system is evaluated using mean Average Precision (mAP), precision–recall, frames per second (FPS), and end-to-end latency from the camera to the dashboard. The prototype achieves an mAP@0.5 of 98.5% with precision–recall above 97%, an average throughput of 8.3 FPS at 640×640 resolution, and a median SSE communication latency of 0.5–0.6 ms, demonstrating the feasibility of a cost-effective solution for automated waste sorting. The system also provides logging, operational statistics, an offline queue, and an idempotency mechanism to support reliable operation in real-world deployments.]]>
Copyrights © 2026
