Multi-camera detection framework for lifelong broiler flock monitoring

Abstract

In recent times, there has been an increased emphasis on chicken welfare within the poultry industry, leading to the adoption of camera-based solutions for continuous monitoring of broilers' activity and health parameters. Accurate detection and localization of broilers within the pen are crucial input for various welfare monitoring applications. However, current research has predominantly concentrated on single-camera setups, which are prone to occlusions caused by feeder or lighting equipment and struggle with scalability in larger pen sizes. To overcome these challenges, we propose a multi-camera setup that allows precise detection and localization of each broiler in a top-down view of the pen. Our approach ensures detection accuracy throughout the entire six-week lifespan of broilers, providing robust and consistent results despite the significant changes in the animals' visual appearance over time. We obtain a MODA score of 83.8% on real-world data and demonstrate its superior effectiveness compared to single-camera setups.