Revolutionizing Cow Welfare Monitoring: A Novel Top-View Perspective with Depth Camera-Based Lameness Classification.

Area of Science:

• Veterinary Medicine
• Agricultural Engineering
• Computer Vision

Background:

• Early detection of lameness in cattle is crucial for animal welfare and farm productivity.
• Existing 2D imaging systems have limitations in accurately assessing lameness-related changes.

Purpose of the Study:

• To develop and validate a novel system for cow lameness detection and classification using top-view 3D depth cameras.
• To differentiate the proposed 3D depth system from traditional 2D methods for improved accuracy.

Main Methods:

• Utilized a 3D depth camera integrated with deep learning for cow detection, tracking, and segmentation.
• Employed Detectron2 framework and Intersection Over Union (IOU) for precise cow localization and movement analysis.
• Extracted depth data, focusing on the backbone region's maximum height for feature vector generation.
• Evaluated classification performance using Random Forest (RF), K-Nearest Neighbor (KNN), and Decision Tree (DT) algorithms.

Main Results:

• Achieved high average detection accuracy of 99.94% and tracking accuracy of 99.92% over a three-day period with varying herd sizes.
• Successfully extracted depth features from the cow's back region for lameness assessment.
• Demonstrated the efficacy of the developed system in classifying lameness using multiple machine learning models.

Conclusions:

• Top-view depth cameras offer a promising, accurate solution for automated cow lameness detection and classification.
• The integrated deep learning approach significantly enhances livestock health management capabilities.
• This technology has substantial implications for improving dairy farm efficiency and animal well-being.