Gait Recognition System Biometric Identification in IoT-Enabled Smart Environments

Abstract

The paper presents a comprehensive IoT-integrated gait recognition system for forensic identity verification in smart surveillance environments. The proposed system uses the OpenGait deep learning framework and two silhouette extraction methods, Robust Video Matting (RVM) and traditional Gaussian Mixture Model (GMM) background subtraction, to process video sequences into normalised silhouette datasets for recognition. The whole pipeline from video ingestion and preprocessing to feature extraction, similarity matching, and delivering the results is orchestrated by a Flask-based web application. The system has two modes, one is for offline forensics of archived videos, and the other is for real-time monitoring using a webcam with MJPEG streaming and ServerSent Events (SSE). Subject profiles are stored in a SQLite database backend with MD5 based de-duplication. Evaluation on controlled indoor datasets demonstrates high identification accuracy and robust performance with small variations in clothing and walking speed. The paper addresses major shortcomings of current gait recognition prototypes such as the lack of acceptable interfaces, integrated databases and live recognition support. Thus the system provides a deployable platform suitable for perimeter security, elder care, retail loss prevention and smart city forensics.