Abstract
A foundational model for monitoring calving in breeding cattle using surveillance video, along with a notification interface for farmers, has been developed and evaluated. The system's effectiveness was assessed through simulation experiments using cattle monitoring footage and real-world deployment on livestock farms. Effective calving assistance requires not only accurate predictions but also an intuitive presentation of the reasoning behind these predictions to support farmers' decision-making. This requirement is crucial not only in designing the notification interface but also in developing the foundational model that underlies the system. To address this, we designed the model to incorporate the decision-making processes involved in calving assistance. Based on this model, we developed a real-time calving alert system that detects signs of calving and presents both predictions and their underlying rationale to farmers. The proposed system's detection performance was evaluated using a cattle tracking dataset, demonstrating superior accuracy compared to existing models that do not incorporate expert knowledge. Furthermore, real-world deployment on livestock farms provided valuable insights into the system's practical utility, as well as the key information and presentation methods necessary for effective decision-making in calving assistance.
