An End-to-End AI-Driven Framework for Automated Missing Person Identification Using Deep Metric Learning and Real-Time Surveillance Integration

Abstract

The paper outlines an AI system that will search and locate missing individuals through a sophisticated face recognition. It relies on deep learning to encode the faces and a scalable web design to ensure that the matching profiles are located instantly between the stored missing-person files and the missing-person reports submitted by the population. The new approach detects with HOG and embeds faces with ResNet 34 to achieve 99.38% accuracy on the LFW test data. The back is built on FastAPI, the front is built on react.js and Docker can be easily deployed. The system is tested to be able to respond to an image query within 2-5 seconds and supports up to 30 seconds to 2 minutes of video with each face match consuming less than 100 milliseconds. It compares the matches based on the Euclidean distance, allows a user to set the threshold, provides automatic alerts via email and displays the location on the Leaflet map. Security applies JWT log in, passwords using bcrypt, and encryption using SHA-256. The system allows it to have over 50 users concurrently and it has a record of over 10,000 active cases that offer a solid foundation on which community-based missing persons searches can be conducted.