Human falls pose serious health and safety risks, especially among the elderly and individuals requiring continuous monitoring. Traditional fall detection methods often rely on wearable sensors or manual supervision, which can be intrusive, expensive, and difficult to scale. This paper presents a real-time fall detection system leveraging YOLOv5nu—a lightweight variant of the YOLOv5 deep learning model—deployed on a Raspberry Pi 4B for efficient edge computing. The system uses computer vision to detect falls based on four key motion descriptors: vertical speed, aspect ratio, orientation angle, and vertical displacement. Upon detecting a fall, it promptly activates a dual-alert mechanism comprising local audio alarms and automated email notifications. Experimental evaluations across diverse indoor and outdoor environments demonstrate a high F1-score of 98.1%, low latency (98 ms), and stable performance without external GPU or cloud support. The system operates entirely offline using low-cost, off-the-shelf components, making it a reliable, scalable, and privacy-preserving solution for smart healthcare applications in hospitals, elderly care facilities, and residential settings.