M-ABCNet: Multi-Modal Aircraft Motion Behavior Classification Network at Airport Ramps

Abstract

Self-driving baggage tractors on airport ramps signify emerging trends for enhancing operational procedures at airports and contribute to the growth of the aviation industry. Airport ramps are characterized by unique mobility requirements, including layout, population, demand, and traffic patterns. Among these, accurate estimation of aircraft movements is paramount for safety and compliance with airport operational regulations. Contrary to the dynamic nature of other environments, airport ramps have predominantly static and slow movements. Even when an aircraft is parked and stationary, other operational vehicles must exercise caution or halt when the airplane is preparing to push back from or into a parking space. This aspect of airport operations has not been addressed adequately in existing research, and prior studies have rarely considered the detection of airplanes on ramps. This work introduces a context-aware multimodal approach for detecting airplane intentions on airport ramps using RGB and thermal cameras. The proposed methodology involves parallel extraction of behavioral features from airplanes and situational context from their surrounding objects. This approach enables estimation of the movement attributes of the aircraft in relation to other objects on the ramp. The effectiveness of this algorithm was validated through a comprehensive dataset collected from the Cincinnati and Northern Kentucky Airport using the proposed platform. Accordingly, the performance improved by 15.76% with the proposed modality through the use of the thermal camera and additionally by 7.29% through utilization of the proposed network.