Multimodal Abnormal Event Detection in Public Transportation

Abstract

This project focuses on improving passenger safety in public transportation systems. As the use of public transport increases, incidents such as passenger fights, theft, vandalism, and fall accidents are also rising. To address this issue, this paper presents a multimodal abnormal event detection system using deep learning. The system uses RGB video, depth data, and audio signals to detect abnormal activities inside public transport vehicles. It is designed to work in autonomous vehicles where no driver is present. Experiments conducted on a custom dataset with events such as fighting, bag snatching, vandalism, and normal behavior show promising results, achieving an overall accuracy of 85.1%.