Dolphin Pool Safety System

The Project

Drowning ranks among the leading causes of injury and death for children in the U.S. My solution, the Dolphin Pool Safety System, proposes a novel approach: a hardware device housing a Raspberry Pi and camera, enclosed in a 3D-printed case, continuously monitors pool activity. The camera feeds footage to an AI model I trained using 500 images and uses CV2 and YOLOv5 for object detection, segmentation, and classification. This model accurately identifies and tracks individuals and pool boundaries by drawing bounding boxes with specific coordinates and alerts a mobile app if the boxes for the pool intersect with the boxes for the children. The app, developed with Flutterflow, offers a simple and user-friendly interface with features such as device pairing, where users can enter a six-digit serial code unique to each device to connect, live pool status updates, where users can see information such as the number of people in the pool, pools detected and if people are in the pool, and alerts on detected individuals in the pool. As this type of product is not commonly seen on the market, my goal is to create a solution that is cheap and readily accessible to the general public.

View this project on YouTube

Team Comments

I chose to develop the Dolphin Pool Safety app to address a critical need for enhanced pool safety and provide peace of mind to pool owners. The inspiration came from dolphins, which are known for their lifesaving abilities and have been documented rescuing drowning humans.

One of the most complex and rewarding parts of developing the Dolphin Pool Safety app was training the AI model using YOLOv5 and integrating it into a functional software solution. This involves finding hundreds of pictures online, annotating them manually and monotonously feeding it to the AI.

In addition to imrpoving the app with a better image quality after annotating and editing using CV2, I would make the app more accessible, including expanding it to smart watches to better alert pool owners. Codingwise, I would instead of monitoring for human, specifically target children.