Human-action recognition aims to identify the actions performed by individuals. Due to the broad spectrum of human activities, action recognition covers a wide range. Among all, fall detection is a critical aspect of surveillance, particularly in environments where individuals are at risk. Throughout the years, several sensors, data types, and classification techniques have been investigated to address this issue. This paper proposes a lightweight fall-detection system that processes image sequences in real time. This system is deployed on an embedded device, specifically the Jetson Nano. Our goal is to construct a comprehensive dataset that accurately detects falls in various lighting conditions. The proposed system is constructed using YOLOv8n-pose, which has been trained to identify people using widely recognized dataset. Our methodology includes the design and implementation of YOLOv8n-pose, data collection, and rigorous testing to ensure the accuracy of real-time fall detection using a surveillance camera. The experimental results show that high detection over 90% accuracy and acceptable timing capabilities are achieved.