Bouy

Bouy detects falls via WiFi signals, so elderly residents need no wearables or cameras, keeping monitoring non-invasive. When a fall occurs, caregivers are alerted and respond from their phone.

Comment

Inspiration

Monitoring elderly relatives typically means asking them to wear a device or accepting cameras in their home. Both create friction and erode dignity. We wanted fall detection that works passively using infrastructure already present in most homes.

What it does

Bouy uses WiFi Channel State Information to detect falls without any wearable or camera. When a fall is detected, the at-risk person gets a response window before their care circle is notified. Alerts go out over WhatsApp and caregivers can respond through a live dashboard showing sensor status, incident history, and escalation options including a 911 prompt if no one responds.

How we built it

Four ESP32 boards measure how WiFi signals change as people move through a room. A trained ensemble of an LSTM and Transformer classifies those changes in real time. The backend runs on Express with SQLite and uses Pusher for real-time updates across two separate PWAs, one for the recipient and one for caregivers, deployed on Railway and Vercel.

Challenges we ran into

Timestamp handling between the ESP32 boards and the server caused stale readings to appear fresh. The fall detection threshold required careful tuning to avoid triggering on normal movement. Keeping the recipient and caregiver views in sync through every phase of an incident without duplicate alerts required tight state machine logic. Data collection was also a lengthy process(falling is not fun).

Accomplishments that we're proud of

End-to-end fall detection from raw WiFi signal to caregiver WhatsApp alert with no wearable involved. The incident state machine handles the full response lifecycle gracefully, giving the person at risk agency before escalating to caregivers.

What we learned

WiFi CSI is a genuinely viable passive sensing modality for indoor fall detection, but the gap between a working model and a reliable deployed system is significant. Sensor freshness, network timing, and user experience under stress all require as much attention as the model itself.

What's next for Bouy

Multi-room coverage, activity pattern baselines to reduce false positives, integration with professional monitoring services, and a native mobile app for caregivers who need reliable background push notifications.

Built With

Share this project:

Updates