Occlusion-Aware Tiny RGB-Thermal Fusion Transformer for Small-UAV Tracking in Cluttered Airspace

سال انتشار: 1405
نوع سند: مقاله کنفرانسی
زبان: انگلیسی
مشاهده: 106

فایل این مقاله در 7 صفحه با فرمت PDF قابل دریافت می باشد

استخراج به نرم افزارهای پژوهشی:

لینک ثابت به این مقاله:

شناسه ملی سند علمی:

UTCONF10_087

تاریخ نمایه سازی: 26 شهریور 1405

چکیده مقاله:

Small unmanned aerial vehicles are difficult to track because their appearance changes across visible and thermal sensors, particularly under low illumination, thermal crossover, and temporary disappearance. This paper presents OATF, a ۴,۷۲۲-parameter fusion transformer that weights multiscale correlation evidence and couples it with a temporally calibrated presence gate. A reproducible study on ۱۴ validation and ۱۸ independent Anti-UAV ۳۰۰ sequences shows that OATF improves average overlap, IOU-۰.۵ success, and normalized precision over the best single modality. However, its official Anti-UAV accuracy remains ۰.۱۵۲ versus ۰.۱۹۶ for infrared-only tracking, exposing absence calibration as the principal limitation.

نویسندگان

Mahdi Galdi Najafabad

PhD in Artificial Intelligence, Professor at Varastegan University of Medical Sciences and Atrak University