Arcle

The Problem

In elderly care facilities, the time between a fall and rescue is critical. Staff cannot continuously monitor every resident, and elderly people often cannot or don't remember to call for help. Existing solutions were either invasive (cameras with continuous recording) or unreliable (emergency buttons that get forgotten or can't be reached after a fall).

I co-founded Arcle as an IoT healthcare startup with the goal of building a system that automatically detected falls without requiring any action from the elderly person, respected privacy, and worked even during internet outages.

The Solution

I designed and developed the entire hardware and software stack. The smart wristbands use an ESP32 with an MPU-6050 3-axis accelerometer and gyroscope: lightweight, battery lasting days, MQTT communication over local network. The detection algorithm analyzes acceleration and rotation patterns in short time windows: it distinguishes a fall from a normal abrupt movement, reducing false positives from 18% to 4%.

The architecture is edge computing by design: Raspberry Pi 4 as the central hub on each floor processes data locally without depending on internet. ESP32 handles distributed environmental sensors (temperature, humidity, air quality) to monitor resident comfort. If the connection drops, the local system continues alerting staff in the facility via LAN.

Cameras in common areas use computer vision to detect postural anomalies (someone on the floor, unusual movements) without continuous recording: real-time processing, no footage saved, only notifications when something is detected.

The Django backend builds a behavioral model for each resident: it learns routines (sleep schedules, habitual paths, activity levels) and detects significant deviations that might indicate a problem. The Flutter app for staff smartphones receives real-time notifications with the resident's location.

The Result

The system was developed and tested with real devices. The edge computing architecture was validated in the field. False positives in fall detection were reduced from 18% to 4%. The project remained at prototype stage: the startup did not proceed toward commercialization.