Semi-Automatic Women Safety System Using Real-Time Facial Distress Detection with Mandatory User Confirmation and Emergency Alert Mechanism

Abstract

Women’s safety remains a critical global concern. Conventional panic applications and wearable devices require manual activation, which is often impossible when the victim is in shock, physically restrained, or under extreme stress. This paper proposes a semi-automatic women-safety mobile system that continuously monitors the user’s facial expressions using a lightweight Convolutional Neural Network (CNN). When a high probability of distress-related emotions (fear, anger, or sadness) is detected for three consecutive frames, the system instantly triggers strong haptic vibration and displays a large full-screen one-tap SOS confirmation button. Only if the user explicitly taps this button within 7 seconds does the system activate a loud deterrent siren and send the current GPS location along with a pre-recorded emergency message to pre-selected trusted contacts and, if the user has opted in during setup, to local emergency services. Experimental results on a combined dataset of approximately 50,000 facial images show a seven-class emotion classification accuracy of 89%. Real-world field trials conducted with 25 female volunteers in public environments recorded zero false or unintended emergency alerts, with an average time from first distress detection to confirmation screen appearance of 6.4 seconds and an average end-to-end alert transmission time of 6.4 seconds (including user confirmation). This is significantly faster than the 15–18 seconds required by traditional manual panic applications, while eliminating the risk of erroneous alerts that would occur in a fully automatic system. The proposed framework offers a practical, privacy-preserving, and ethically responsible solution that can be readily deployed on existing smartphones and wearable devices, contributing meaningfully to AI-driven personal safety technologies.