Cardoen, ThorstenThorstenCardoende Carvalho, Patricia SosterPatricia Sosterde CarvalhoAntonissen, GuntherGuntherAntonissenTuyttens, Frank A. M.Frank A. M.TuyttensLeroux, SamSamLerouxSimoens, PieterPieterSimoens2025-05-272025-05-272025-05-2720250168-1699WOS:001493191500003https://imec-publications.be/handle/20.500.12860/45726Welfare concerns in poultry farming have driven the need for advanced monitoring solutions to study broiler activity and health. However, existing research predominantly relies on single-camera setups, which are prone to occlusions from equipment such as feeders and lighting, limiting their effectiveness. To address this, we propose a multi-camera setup that enables comprehensive broiler localization and tracking from a top-down view of the pen. To support this approach, we introduce MVBroTrack,1 an open-source dataset containing real-world data with annotations for various subtasks critical to broiler studies. We demonstrate robust performance of our multi-view detection pipeline throughout the six-week broiler lifespan despite significant changes in visual appearance. Additionally, we present a novel unsupervised tracking method that surpasses the traditional tracking by detection paradigm, improving the IDF1 score by 3% and increasing the proportion of mostly tracked broilers by 5%. By reducing the need for manual observation, our multi-camera pipeline facilitates exhaustive studies of broiler behavior and welfare, paving the way for significant advancements in poultry research and farming practices.Journal article10.1016/j.compag.2025.110435WOS:001493191500003