Utilizing Visual-Spatial Perception to Detect Drowning-induced Suffocation in Swimming Pools

Drowning is the third leading cause of unintentional injury-related death worldwide. In the US, drowning is responsible for upwards of 4,000 deaths annually, with young children being particularly vulnerable. Traditional methods of drowning detection, such as lifeguards and surveillance cameras, are ineffective or too expensive to be implemented in many areas. This paper presents a low-cost approach to drowning detection using artificial intelligence-based object detection, object tracking, and Pose estimation on a low-cost single-board computer device (Raspberry Pi) with a camera module. The engineered system consists of a Raspberry Pi-based device utilizing a camera, a deep learning model for object detection, tracking, pose estimation, and a custom-designed algorithm for drowning detection. The deep learning model is trained on a large dataset of video frames of people swimming, near drowning, and drowning. The custom-designed algorithm running on the device utilizes the output of the deep learning model to track individuals in the water and identify potential drowning victims. The device was evaluated in a controlled environment and achieved an accuracy of $\mathbf{88\%}$ in detecting drowning. Additionally, it can track individuals in real time and identify potential drowning incidents within a few seconds. The device can be used in various settings, such as public and private pools to prevent drowning. The system is low cost and easy to implement, making it a viable option for many communities.