Tracking and Behavior Analysis of Group-Housed Pigs Based on a Multi-Object Tracking Approach.

Area of Science:

• Agricultural Technology
• Animal Behavior Science
• Computer Vision

Background:

• Monitoring pig health and welfare in group-housed settings is crucial for smart farming.
• Traditional methods struggle with challenges like variable lighting, occlusion, and pig clustering, leading to tracking errors.
• Accurate, automated behavior tracking is needed to improve livestock management and early disease detection.

Purpose of the Study:

• To develop a robust multi-object tracking (MOT) approach for automatic monitoring of group-housed pig behaviors.
• To address common challenges in pig tracking, including variable lighting, occlusion, and pig clustering.
• To provide a reliable solution for real-time behavior tracking, enhancing pig health and welfare management.

Main Methods:

• Utilized YOLOv8 for real-time pig detection and behavior classification under challenging conditions.
• Employed OC-SORT for robust multi-object tracking, effectively handling pig clustering and occlusion.
• Developed an automatic behavior analysis algorithm integrating detection and tracking data for long-term monitoring.

Main Results:

• The V8-Sort method achieved superior performance compared to JDE, Trackformer, and TransTrack on pig tracking datasets.
• Achieved high HOTA, MOTA, and IDF1 scores (e.g., 82.0%, 96.3%, 96.8% on 1-minute videos).
• The system accurately recorded durations of specific pig behaviors, providing insights into health and welfare.

Conclusions:

• The proposed YOLOv8 + OC-SORT (V8-Sort) approach offers excellent performance in pig behavior recognition and tracking.
• This technology provides essential technical support for anomaly detection and health status monitoring in pig farming.
• The system contributes to improved animal welfare and efficient management in smart farming environments.