Oriented bounding boxes improve aerial vehicle localization accuracy
OBB-PBD: Extending Parallel Box Decoding to Oriented Bounding Boxes for Referring Expression Grounding in Aerial Imagery
Computer Vision and Pattern Recognition
Summary
Finding objects like cars or trucks in aerial images is tricky because they can be turned in any direction. The authors improved a method that guesses where objects are by allowing it to draw rotated boxes instead of just upright rectangles. They tested their new approach and showed it works better at marking these objects' tilted positions. This helps make identifying and locating objects from above more accurate.
What this means in practice
- •For aerial imagery analysts: Enhance detection and localization of aerial vehicles by providing more precise rotated bounding boxes.
- •For drone operations teams: Improve automated tracking of oriented objects in overhead drone footage for situational awareness tasks.
Authors
Chang Woo Song, Yeong Jun Koh
Abstract
Referring expression grounding in overhead imagery requires oriented bounding boxes rather than horizontal ones, since aerial objects appear at arbitrary orientations. Most grounding models emit box coordinates sequentially; Parallel Box Decoding (PBD) instead predicts all geometry slots in one forward pass, but only for four-coordinate horizontal boxes. We extend this block-aligned interface to a six-slot oriented representation using a double-angle parameterization. Rather than assuming correctness from a decreasing loss curve, we verify the extension with image/prompt-shuffle controls and visibility/leakage probes, which show that the corrected model depends on both the image and the prompt rather than on prompt statistics alone. On the aerial vehicle visual grounding (AVVG) benchmark, whose targets are genuinely oriented, the extended model improves rotated intersection over union (rotIoU) from 0.1253 for the zero-shot baseline to 0.303 +/- 0.012 (mean +/- standard deviation over three training seeds) across the complete 6139-record test split spanning all 120 images. Learning oriented boxes rather than their axis-aligned envelopes adds 0.055 +/- 0.008 rotIoU beyond a matched control (per-seed 95% confidence intervals all excluding zero), whereas block expansion alone yields no rotIoU gain distinguishable from zero.