Person Re-Identification for Public Safety in Indian Railways using deep learning

Abstract

Ensuring public safety in large-scale environments like Indian Railways requires robust surveillance mechanisms. Traditional systems relying on handcrafted features suffer from low accuracy due to sensitivity towards variations in lighting, pose, and camera angles. To address this limitation, a deep learning-based approach for person re-identification is proposed. The system employs Convolutional Neural Networks (CNNs) for automatic feature extraction from video feeds, followed by classification using machine learning algorithms such as Random Forest and Support Vector Machines (SVM). The architecture supports dual interfaces for administrators and railway employees, facilitating real-time monitoring and alert generation upon detecting matches with a database of suspicious individuals. Experimental evaluation indicates that Random Forest achieves higher accuracy compared to alternative classifiers, making it suitable for deployment. Comprehensive testing validates system functionality, while challenges related to manual surveillance and data acquisition are acknowledged. Future enhancements include integration with live CCTV feeds and scalability improvements.