Embedded Machine Learning-Powered Assistive Wearable for Fall Detection and Safe Navigation

This paper investigates a dual-purpose smart assistive technology framework using Tiny Machine Learning (TinyML) to support particular groups of population, specifically the elderly and visually impaired.The suggested system combines two separate but compatible modules: a blind visual assistance device and a fall detection system. Both are driven by small, energy-efficient machine learning models that are installed on single board computers (SBC) and low-power microcontrollers. The fall detection component takes on the task of motion sensors such as accelerometers to constantly monitor body movement and orientation. This triggers immediate alerts to guardian upon detecting a fall event. Concurrently, the blind assistance system module uses ultrasonic sensors, a camera module and voice feedback to enable real-time obstacle detection and distance estimation. Both systems are designed for wearable application and processing is done locally. This makes the system portable, lower latency and lower power consumption. For older and visually impaired people, the combined framework shows how machine learning models and IoT can provide affordable, real-time and highly accurate assistive. This approach greatly increase safety and independence for a challenged person. Keywords Deep Learning, Object Detection, Fall Detection, IOT, SBC, Embedded Systems