Detection and Classification of Abnormal Passenger Behaviour in Self Driving Buses Using In-Vehicle Camera Systems

With the rapid advancement of self-driving technology in public transportation, ensuring passenger safety remains a critical challenge. Identifying and responding to unsafe or inappropriate passenger behaviors is essential, particularly in autonomous vehicles operating without direct human oversight. This study presents a novel approach for detecting and classifying abnormal passenger activities within a bus environment. Unlike conventional human activity recognition methods, the proposed system utilizes an overhead vision-based framework to reduce occlusion and improve detection accuracy. A specialized action recognition network is designed to process top-view images, effectively capturing both spatial and temporal dynamics for enhanced classification. To facilitate real-world implementation, a dedicated dataset, BUS-HAR, has been developed, containing diverse activity samples for robust model training. Experimental evaluations on real-world data demonstrate the superior performance of the proposed method over existing techniques, highlighting its potential for improving safety in autonomous public transport.