Real-Time Abnormal Event Detection for Enhanced Security in Autonomous Shuttles Mobility Infrastructures.

Area of Science:

• Computer Science
• Artificial Intelligence
• Transportation Engineering

Background:

• Autonomous vehicles (AVs) are increasingly deployed globally, raising concerns about their societal impact and safety.
• The absence of human drivers in AVs necessitates new security measures, especially in urban environments with heightened terrorism threats.
• Traditional driver-led incident management for passenger behavior and petty crimes is no longer applicable in AVs.

Purpose of the Study:

• To address the security challenges in autonomous shuttle mobility infrastructures.
• To develop an automated system for detecting abnormal passenger behavior and petty crimes in AVs.
• To enhance passenger security and the overall safety of autonomous transportation systems.

Main Methods:

• An online, end-to-end deep learning solution was developed for real-time detection.
• The system utilizes camera sensor data and smart software for analysis.
• The deep learning model was trained to identify various types of petty crimes and abnormal behaviors.

Main Results:

• The proposed system accurately and robustly detects diverse petty crime types, including aggression, bag-snatching, and vandalism.
• It effectively identifies abnormal passenger behavior such as vandalism and accidents.
• The solution demonstrates excellent performance across various use cases and environmental conditions.

Conclusions:

• Deep learning-based surveillance systems can significantly enhance security in autonomous vehicles.
• Automated detection of abnormal behavior and petty crimes is crucial for the safe integration of AVs.
• The developed system offers a viable solution for improving passenger safety in autonomous shuttle services.