Edge AI-Driven Multi-Modal Silent Distress Detection for Domestic Abuse Victims Using TinyML and Sensor Fusion

Victims of domestic violence frequently endure in silence due to fear, compulsion, or persistent surveillance, inhibiting their ability to seek assistance. This study introduces an innovative Artificial Intelligence of Things (AIoT) platform aimed toward discreet emergency detection in smart home settings. The suggested system incorporates non recording microphones, passive infrared (PIR) motion sensors, and ambient environmental detectors to observe behavioral and environmental indicators while safeguarding user privacy. Employing Tiny Machine Learning (TinyML) on ESP series microcontrollers facilitates fast on device inference independent of cloud infrastructure, thereby guaranteeing low latency and data security. A bespoke Long Short Term Memory (LSTM) network analyzes sequential multi modal sensor data to detect nuanced discomfort patterns indicative of suspected abuse. Moreover, end to end encryption procedures ensure that distress warnings are securely conveyed to authorized recipients. The amalgamation of sensor fusion with edge computing improves precision, reactivity, and scalability while upholding robust privacy protections. Experimental validation in simulated domestic contexts exhibits elevated detection accuracy, negligible false alarms, and swift response times. This work presents a privacy preserving, adaptive, and scalable AIoT solution that enhances real time, discreet intervention technologies, enabling smart home safety systems to offer dependable protection for vulnerable individuals in abusive circumstances.