Multiview Monitoring of Individual Cattle Behavior Based on Action Recognition in Closed Barns Using Deep Learning.

Area of Science:

• Animal Science
• Computer Vision
• Agricultural Engineering

Background:

• Cattle behavior recognition is vital for animal welfare and health monitoring.
• Current methods face challenges with individual identification due to occlusion and pose variations.
• Accurate individual tracking is crucial for reliable behavior analysis in livestock.

Purpose of the Study:

• To develop a multiview monitoring system for recognizing individual cattle behavior.
• To address challenges in identifying and tracking individual cattle in farm environments.
• To enable continuous monitoring and statistical analysis of cattle behavior.

Main Methods:

• Utilizing video data sequences as input for action recognition.
• Employing a detector to identify hierarchical actions (part and individual actions).
• Implementing a tracking and identification mechanism for continuous individual monitoring.

Main Results:

• The system successfully tracks and assigns unique identification numbers to individual cattle.
• Demonstrated effectiveness through quantitative and qualitative experimental results on a Hanwoo cattle database.
• Captured spatiotemporal information for automatic cattle behavior recognition.

Conclusions:

• The proposed framework effectively tackles real-world farm indoor challenges for cattle behavior recognition.
• Enables automatic recognition of individual cattle behavior, supporting precision livestock farming.
• Provides a robust solution for monitoring cattle health and welfare through behavior analysis.