Skip to content
Open access

OcluTrack: A Persistent Identity Tracking Framework for Individual Poultry Monitoring Under Dynamic Occlusion

2026 · IEEE Access · Vol 14, pp. 163417-163429 · 0 citations · 36 references

Abstract

Individual monitoring of commercial poultry (laying hens and broiler chickens) is essential for welfare assessment and early detection of health anomalies. Yet, it remains technically challenging due to low lighting, high bird density, high phenotypic similarity, and frequent dynamic occlusion among individuals. This paper presents OcluTrack, an end-to-end computer vision pipeline that integrates bird detection with a compact YOLO model and persistent identity tracking through a virtual marker (ghost) mechanism. The system assigns stable identifiers (PIDs) to each bird throughout the entire recording session, even during total occlusion events, without relying on appearance-based re-identification networks. Re-identification is performed by global optimization via the Hungarian Algorithm with a combined score (multipoint distance, IoU, and size ratio), complemented by multi-layer quality strategies: confirmation hysteresis, boundary rejection, noisy zone blocklisting, and a static artifact detector. Validation experiments were conducted in an experimental production environment with high bird density, visually homogeneous individuals, low lighting, and a high occurrence of detection noise. Nine birds were simultaneously tracked in ten-minute sessions. OcluTrack reduced the unique identifiers of the baseline trackers ByteTrack and BoT-SORT from 66 to 9 (actual number of birds) in both evaluated videos, with zero ID-switches across all variants. The computational overhead on the baseline tracker was + 1.84%, maintaining processing above 40 FPS. The results indicate that the integration of lightweight detection and analytical persistent tracking presents lower trajectory fragmentation and higher identity persistence compared to the evaluated trackers, demonstrating viability for edge computing.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.