Methods and challenges in computer vision-based livestock anomaly detection, a systematic review

With the rapid advancements of livestock production, computer vision has attracted considerable attention as a non-invasive and effective technology for detecting livestock anomalies. Despite its potential, challenges persist in complex scenarios, such as the difficulty of distinguishing between normal and abnormal states, the inherent complexity and variability of anomalies, and environmental factors that contribute to missed detections and false positives. This paper provides a comprehensive review of the development and applications of computer vision in livestock anomaly detection over the past six years. First, the current research status in this field is introduced, analysing the collected literature by species, countries, years, and equipment applications. Next, various types of anomalies and their corresponding detection methods across spatial, temporal, and spatiotemporal dimensions are categorised. Finally, an in-depth discussion of challenges related to data, task, and models is offered to propose practical solutions to these challenges, and highlight promising directions for future research. We recommend fostering collaboration among multidisciplinary researchers to address challenges in detecting, identifying, and tracking individual anomalies in livestock. These challenges include the difficulty of collecting anomalous samples, the uncertainty and variability of anomalies, and the high similarity among livestock, all of which hinder effective long-term monitoring.