BOVIDS

BOVIDS employs deep learning to identify and classify nocturnal behavioral poses in African ungulates from video and image data to enable quantitative analysis of nocturnal activity patterns.

Key Features:

• Pose estimation: Identifies three behavioral poses — "Standing", "Lying-head up", and "Lying-head down".
• Classification accuracy: Achieves 99.4% accuracy for the specified pose classes.
• Data processing scale: Applied to over 11,411 hours of video footage in the described case study.
• Case study dataset: Analyzed observations from 25 common elands (Tragelaphus oryx) across five EAZA zoos covering 822 nights.
• Analytical outputs: Produces quantitative estimates of nocturnal activity budgets, phase lengths, and number of phases per behavioral state.

Scientific Applications:

• Nocturnal behavior quantification: Enables detailed description of nightly behavior in common elands and other African ungulates from video material.
• Comparative demographic analysis: Supports analyses of age- and sex-dependent differences in nocturnal activity budgets, phase lengths, and phase counts.
• Sleep posture analysis: Detects differences in time spent in REM sleep posture between males and females and across age classes.
• Behavioral rhythm detection: Facilitates identification of rhythmic patterns between Standing and Lying phases relevant to behavioral ecology.
• Welfare and husbandry assessment: Provides quantitative data to inform animal welfare assessments and optimization of husbandry conditions.

Methodology:

BOVIDS employs deep learning algorithms to automate the identification and classification of behavioral poses from video and image data.