Objective dairy cow mobility analysis and scoring system using computer vision-based keypoint detection technique...

Area of Science:

• Animal Science
• Veterinary Medicine
• Computer Vision

Background:

• Assessing dairy cow mobility is crucial for animal welfare and productivity.
• Traditional visual scoring can be subjective and labor-intensive.
• Objective methods for mobility assessment are needed.

Purpose of the Study:

• To evaluate a computer vision technique for extracting mobility variables from 2D videos of dairy cows.
• To develop and assess a machine learning model for predicting dairy cow mobility scores using these variables.

Main Methods:

• Collected 256 top-view 2D videos of walking dairy cows.
• Used a keypoint detection model to track 10 keypoints and extract 25 mobility variables.
• Developed a random forest classification model for 3-level mobility scores (0, 1, 2+3).
• Evaluated model performance using repeated holdout (80% train, 20% test) with weighted kappa and AUC.

Main Results:

• The computer vision technique effectively extracted relevant mobility variables.
• The random forest model achieved a weighted kappa of 0.72 and an AUC of 0.89 for 3-class mobility prediction.
• Key variables for prediction included back lateral movement, withers lateral movement, walking speed, and tail head lateral movement.

Conclusions:

• Computer vision-based keypoint detection is a viable method for extracting mobility variables from dairy cow videos.
• Machine learning models utilizing these variables show strong potential for objective and accurate mobility scoring in dairy cows.
• This approach can aid in early detection of mobility issues and improve herd management.