<![CDATA[Falls in the elderly represent a serious global health problem. According to the WHO, one in three elderly people aged over 65 years experiences a fall each year. In Panti Werda Kediri, monitoring of elderly activities is still performed manually by staff, causing fall incidents to go undetected quickly. Previous YOLO-based fall detection studies generally produce models without integrating them into monitoring platforms usable by non-technical end users and without automatic notification. This research aims to determine the effectiveness of a monitoring system in detecting normal activities and fall incidents in elderly residents in real time at Panti Werda Kediri using YOLOv8. The system was developed using the Waterfall method through stages of requirements analysis, system design, implementation, and testing. The detection component uses a retrained YOLOv8 model to recognize two classes: normal and fall. The backend is built with FastAPI and PostgreSQL, equipped with a web-based monitoring dashboard and automatic notifications via Telegram Bot. A fall confirmation mechanism based on 3 consecutive frames with a 1.5-second cooldown suppresses false positives. Blackbox testing conducted at Panti Werda Kediri shows all 10 test scenarios passed. The system successfully sends real-time Telegram notifications in under 2 seconds with visual evidence each time a fall is confirmed, provides live camera streaming, and displays complete detection history through a web dashboard accessible to non-technical staff.]]>
