<p>Drowning remains a critical public health issue, particularly among children and young adults, often due to delays in recognizing and responding to emergencies. There are many previous methodologies for drowning detection that have predominantly utilized conventional techniques, including human observation, closed-circuit television monitoring, and rudimentary alarm systems. Despite their widespread implementation, these approaches are often limited by critical limitations such as susceptibility to human error, observer fatigue, and latency in incident recognition. To address these limitations, this study proposes an intelligent drowning detection system utilizing the YOLOv11 model, optimized for real-time, low-power operation on a Raspberry Pi 5 (8GB RAM). A custom dataset was developed to represent different real-world drowning scenarios. The dataset was prepared through cleaning, annotation, and augmentation, followed by model training. To improve model robustness and generalization, augmentation techniques such as scaling, rotation, brightness adjustment, and noise reduction were applied. The proposed system continuously analyzes live video streams to detect potential drowning incidents. Upon detection, the system immediately activates an alarm and sends mobile alarms to lifeguards or emergency personnel, enabling prompt intervention. The model achieved a high mAP@50 score of 0.98 while maintaining a lightweight architecture with 2.58 million parameters and 6.3 GFLOPs (Giga Floating Point Operations Per Second). Due to its affordability, efficiency, and strong performance, the system is well-suited for deployment in swimming pools, beaches, and water parks, providing a practical solution for enhancing safety and facilitating rapid emergency response.</p>