Work

FlowReID - An integer programming approach to re-identification in livestock tracking using quantum and classical optimization

Year: 2026

Type: article

Abstract: Accurate tracking of individual animals is essential for precision livestock farming, enabling health monitoring, behavior analysis, and phenotyping. Multi-object tracking in livestock environments is challenging due to frequent occlusions, close interactions, and visually similar animals. In this study, we present FlowReID, a targeted post-processing re-identification (reID) approach that extends the BoT-SORT tracking framework to make it suitable for optimization using quantum computing. For this, we formulate the reID problem as a binary linear program and subsequently reformulate it as a Quadratic Unconstrained Binary Optimization (QUBO), enabling optimization using a classical solver, simulated annealing, or quantum annealing. Evaluation on video sequences of 16 Holstein Friesian cows from eight camera views demonstrated that FlowReID consistently improved identity consistency compared to the standard Ultralytics reID module, reducing the number of identity switches from 9 to 5, while improving overall multi-object tracking metrics like HOTA (89.7 to 90.8), MOTA (95.4 to 96.3), and IDF1 (96.0 to 97.4). Although quantum solvers were evaluated, they did not outperform classical methods in either speed or accuracy under current quantum hardware constraints. While the proposed algorithm is quantum-ready from a formulation perspective, we did not observe a quantum advantage. Nevertheless, this work provides a proof-of-concept for implementing multi-object tracking in precision livestock farming in a quantum-compatible optimization framework. This formulation enables direct transfer to future quantum or hybrid solvers as quantum hardware continues to mature. Since FlowReID is also implemented using classical optimization solvers, our newly developed approach can be deployed without requiring access to quantum computing hardware.

Source: Smart Agricultural Technology

DOI: 10.1016/j.atech.2026.102498