Dairy cows usually rest for over 10 hours each day, making it crucial for their health, well-being, and productivity to have a dry, clean, and comfortable resting area, such as a freestall. A significant factor affecting the comfort of these stalls is how easily cows can rise and lie down. Therefore, farm employees often monitor unusual rising behaviors as part of regular welfare management.
Dairy cows usually rest for more than 10 hours daily, necessitating a clean, dry, and comfortable place—like a freestall—for them to lie down and rest. This is vital for their health, well-being, and productivity. One important aspect of stall comfort is how effortlessly cows can get up and down. As such, farm workers often look out for unusual rising behaviors as part of their regular welfare management. A recent study published in the Journal of Dairy Science by researchers from Sweden, alongside Sony Nordic, has introduced an automated model that accurately identifies posture changes in dairy cows. This groundbreaking method employs 3-dimensional (3D) pose estimation to provide meaningful, unbiased insights into animal welfare, potentially offering a less time-intensive and more reliable assessment tool for both researchers and farmers.
The research, led by Niclas Högberg, DVM, and Adrien Kroese, Eng, from the Department of Clinical Sciences at the Swedish University of Agricultural Sciences in Uppsala, aimed to create a dependable method to monitor how easily cows can rise and recline in their stalls, which is a key indicator of their overall comfort and welfare.
Adrien Kroese mentioned, “There is significant evidence linking limited movement in cows to decreased welfare. Therefore, various observation practices are common to detect signs of mobility issues.”
Conventional methods of monitoring often depend on human observation, which can be subjective, irregular, and labor-intensive.
In response to the need for more consistent monitoring techniques, the research team proposed an innovative system to track cow movements, specifically focusing on measuring the lying-to-standing transitions utilizing 3D pose estimation data against human observations.
The team set up seven cameras to monitor a herd of Swedish Holstein and Swedish Red cows over a 24-hour period. The recorded video footage was analyzed using 3D pose estimation software, which tracks movements through a 2D object detection and pose estimation system. This data is then processed through convolutional neural networks to recognize cow movements in relation to specific anatomical points observed in static images from the footage, resulting in a 3D representation of cow movements in their stalls, pinpointing the actions that signify a transition to standing.
Kroese elaborated, “We compared the standing data generated by the software with timestamps in the videos that had been annotated by three human observers, which is the recognized standard for behavioral monitoring.”
How did the 3D data model compare to human observations? Kroese stated, “The framework was adept at identifying when a cow moved from lying to standing with accuracy comparable to human observers, achieving a detection sensitivity of over 88%.”
Interestingly, the results showed that the model did not introduce more bias than human observers.
Despite its limitations, the study highlights the potential of 3D pose estimation technology to yield objective and reliable data regarding cow behavior. Kroese emphasized, “This technology marks a significant step forward in our capacity to study and track animal behavior and welfare. By automatically and precisely detecting posture changes, we can acquire valuable insights into the comfort and welfare of dairy cows.”
The model holds promise for researchers looking to expand their studies on dairy cow behaviors and movement patterns and lays the groundwork for new assessment tools that can aid farmers in making informed decisions regarding their herds.
