AI-Driven Real-Time Threat Detection System for Women’s Safety Using Deep Learning and Gesture Analytics
Authors:-Associate Professor Dr.S. Mohana, Preethi S, Sujay Charan P, Parthiban S, Sanjai Krishnan A, Pradiksha R J
Abstract-This paper presents a comprehensive AI-driven real-time threat detection framework designed to enhance women’s safety in urban environments. The system integrates advanced computer vision techniques including YOLOv8- based person detection, ResNet-50 gender classification, and LSTM-based gesture recognition to analyze live surveillance feeds. Through the execution of a multi-modal threat analysis algorithm, the system can recognize vital scenarios including SOS gestures (with 92.3% accuracy), isolated female detection in night hours (with 89.7% accuracy), and probable mob scenario situations. The system is integrated with a distributed architecture supporting low- connectivity location-based edge computing, real-time alert generation based on Twilio/Vonage APIs, and compatibility with police systems through an exclusive web dashboard. Experimental outcomes prove 86.4% threat detection accuracy at an average latency of 1.2 seconds on NVIDIA Jetson devices.
