Autonomous robot for fall detection and emergency alerts

Elderly people suffer more frequent and more severe falls as they age. To ensure swift treatment for their injuries, I invented a robot that can navigate autonomously with sensors collecting spatial data. Based on a neural network, I trained a computer vision model that enables the robot to detect falls. The robot’s alert system allows it to send emails with attached pictures of the person lying on the floor. The main hardware components comprise a Raspberry Pi 4B+ (microprocessor), an Arduino Uno (microcontroller), a camera, four motors, two ultrasound sensors, and a keypad. The robot uses the “Bookworm” operating system and is mainly programmed in Python. 

The main results are that the robot correctly identifies “lying people” with an accuracy of over 80%, and that “walking” as well as “chair-sitting” people are correctly classified as “not-lying.” However, problems in identification seem to arise when people are sitting on the floor.

Beyond better hardware components, possible improvements include methods to autonomously follow the person instead of searching the entire apartment and additional fall detection systems e.g., using an accelerometer. Incorporating a language processing model and enabling live streaming of relevant images could facilitate an initial medical assessment of the situation.