Inspiration
A fall can leave someone unable to reach their phone. We wanted to give people who live alone a way to alert someone they trust using a webcam and a PC.
What it does
FAAALL detects a possible fall and plays a voice alert. The person has 10 seconds to cancel by default. If the countdown runs out, FAAALL sends a Telegram message to a trusted contact with the time and address. The contact can reply “I’ll be there in 10 minutes,” and FAAALL displays their estimated arrival time on the PC.
How we built it
We built a Python desktop app for Windows. OpenCV captures webcam footage, and MediaPipe estimates body pose locally. FAAALL looks at how that pose changes over time to detect a possible fall. We use eSpeak NG for voice alerts, the Telegram Bot API for messaging, and OpenAI to interpret the contact’s reply.
Challenges we ran into
People bend over, sit down quickly, and lie down on purpose. We had to distinguish those movements from falls, even when someone was partly out of frame. We also had to match each Telegram reply to the right alert and prevent duplicate messages when the network timed out.
Accomplishments that we're proud of
We connected webcam detection, a voice countdown, and a real Telegram alert in one Windows app. We confirmed that a webcam-triggered alert reached Telegram and tested that a reply could update the correct incident on the PC.
What we learned
A fall alert is a two-way conversation. The person near the PC needs to know whether someone is coming and how soon. We also learned how much camera placement, occlusion, and ordinary movements affect detection.
What's next for FAAALL
We want to test in more rooms and lighting conditions, measure missed falls and false alarms, and verify the full Telegram round trip on a locked iPhone. We also want to make setup clearer so users can tell when their alert channel is ready.
