Vision-Based Vehicle State and Behavior Analysis for Aircraft Stand Safety

Area of Science:

• Computer Vision
• Artificial Intelligence
• Aerospace Safety

Background:

• Current aviation safety standards necessitate precise monitoring of ground support vehicles in aircraft stand areas.
• Existing surveillance methods like surface movement radar and multi-camera systems have limitations in cost, deployment, and coverage.
• A need exists for a cost-effective and efficient solution for vehicle state perception and behavior analysis in aircraft stands.

Purpose of the Study:

• To propose a lightweight, vision-based framework for vehicle state perception and spatiotemporal behavior analysis.
• To enhance aircraft stand operational safety by accurately monitoring ground support vehicles.
• To leverage existing monocular surveillance resources for improved safety monitoring.

Main Methods:

• Developed a framework utilizing self-calibration and homography transformation for pixel-to-physical plane mapping.
• Integrated an improved lightweight YOLO detector (with Ghost modules and CBAM) and ByteTrack for robust vehicle trajectory extraction.
• Constructed a semantic map and employed a spatiotemporal finite state machine for behavior analysis, fusing position, zone, and time constraints.

Main Results:

• Achieved an average physical localization error of 0.32 m (RMSE).
• The enhanced detection model reached 90.4% accuracy (mAP@50) for ground support vehicles.
• Demonstrated high recall (96.0%) and precision (95.8%) for detecting 'area intrusion' and 'abnormal stop' violations.

Conclusions:

• The proposed vision-based framework is effective and accurate for monitoring ground support vehicles and analyzing their behavior.
• The system offers a cost-effective and easily deployable solution by utilizing existing surveillance infrastructure.
• This framework has the potential to significantly augment current airport ground operation safety monitoring systems.