Wide-Area Motion Imagery Vehicle Detection in Adverse Conditions
Miles Cobb, Matthew D. Reisman, Paul S. Killam, Gawan Fiore, Raasin Siddiq, Danny Giap, Gary Chern · 2023
Automated extraction of intelligence from wide-area motion imagery (WAMI) is a known challenge in the artificial intelligence and computer vision fields, with vehicle detection in particular providing wide-ranging applications to defense, environmental awareness, and economic monitoring. To properly trust automated WAMI image processing methods for remote sensing and situational awareness, a thorough investigation of where and why their performance deteriorates is critical. In this work, we explore boundary conditions of WAMI vehicle detection to identify situations where additional data can address existing performance degradation. Additionally, we propose a method of weakly-supervised detection to reduce reliance on fully labeled ground truth.