Low-cost omnidirectional tracking improves small object monitoring
Re-engineering SORT-based algorithms for low-cost small object tracking from omnidirectional footage
Computer Vision and Pattern Recognition
Summary
Tracking several moving objects on video is tricky when the camera view is wide and objects are tiny and fast. The authors improved a popular tracking method called SORT to handle the special camera view that wraps around, stopping errors where image boundaries meet. They tested their changes on a new video dataset of flying animals and showed better tracking without needing complex appearance analysis. Their approach works well even with simple, CPU-only computing.
What this means in practice
- •For wildlife monitoring teams: Improve tracking of small, fast animals captured by omnidirectional cameras in remote environments without needing high-cost hardware.
- •For urban surveillance operators: Deploy low-cost omnidirectional cameras with enhanced tracking accuracy for crowded scenes without relying on expensive appearance-based models.
Authors
Xin Shu, Meegan Gower, Yvonne Buckley, Anil Kokaram
Abstract
Multi-object tracking (MOT) has advanced rapidly in urban surveillance and autonomous driving, yet many trackers rely on ReID- and transformer-based appearance encoders and are designed for standard FoV cameras. These assumptions break down for low-cost omnidirectional deployments, where equirectangular projection introduces seam discontinuities and targets appear to be small and fast-moving. We address multi-object tracking of flying animals captured in remote environments using omnidirectional cameras. We propose a lightweight framework that re-engineers SORT-based tracking for this geometry, including (i) a Seam-Aware Motion Model that keeps the Kalman state continuous across the seam, (ii) a composite seam-aware association cost that pairs a wrapped Euclidean term with GIoU, and (iii) OmniSmall, a new benchmark of omnidirectional wildlife footage. On our new dataset, with ground-truth detections, our modifications improved over OCSORT by +8.51 HOTA, +9.41 MOTA, and +10.17 IDF1; with YOLOX detections the gain narrows to +1.95 HOTA. Our proposed methods improved tracking performance on OmniSmall and remained competitive on JRDB without adding appearance encoders while keeping the tracking stage CPU-only. Our dataset and source code are available at: https://github.com/Xin-Shu/OmniSORT.git.