Answers are grounded in this paper and cite their sources.
Demonstrates effective video abstraction in monitoring livestock welfare, suggesting advancements in pig management technology.
Monitoring pig welfare and behavioral performance is a fundamental need in precision pig management. Although camera systems are widely installed in pig production facilities, monotonous visual scenes and subtle behavioral changes remain challenging for both manual monitoring and computer vision-based analysis. This study introduced a novel unsupervised video abstract model, named Depth Enhanced Unsupervised Video Abstract (Depth-UVA) model, for trimming long surveillance videos of gestation pigs to condensed videos. The Depth-UVA model integrates unsupervised temporal action segmentation with monocular depth estimation to jointly capture temporal structure and salient spatial regions, enabling comprehensive video content analysis. when evaluated against manually annotated pig posture data, the Depth-UVA model achieved an average 0.93 Mean over Frames (MoF) and 0.90 Intersection over Union (IoU) in posture-related video segmentation. Notably, the model could preserve rare or subtle pig-behavioral and human-related events. By adjusting visual sensitivity and visual target parameters, the Depth-UVA model allows flexible control over abstraction granularity and attention region emphasis, enabling video summaries tailored to different monitoring objectives. This work represents a significant advancement in unsupervised video processing for livestock surveillance and demonstrates practical potential for reducing labor demands, improving management decision-making, and accelerating the adoption of artificial intelligence in Precision livestock farming (PLF).
No takes yet. Share an insight, caveat, or question.
Zhou et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: