Advancing Dairy Locomotion Analysis with Biomechanical Insights from Kinetic and Kinematic Measurements
Authors: Junsheng Zhu and Shabnaz Mokhtarnazif and Gabriel M. Dallago and Abdoulaye M. Diallo and Elsa Vasseur and Kevin Wade
Date: 2026-05-01
Journal: Research Square
Status: Published
DOI: https://doi.org/10.21203/rs.3.rs-9708073/v1
Lameness is a significant problem on numerous dairy farms, jeopardizing both profitability and animal welfare. In this study, kinetic measurements (examining the pressure distribution images of four hooves while a cow is standing statically on a pressure map) and kinematics (analyzing body markers of walking cows) were used to further understand this illness condition. The kinetic data contained 66 records measured 3 times from twenty-three cows (3 missing records) with binary target labels including 8 lame cases and 58 sound cases. Kinetic pressure images were used to extract variables and transform them by asymmetric indices with several machine-learning algorithms. For the asymmetric transformed dataset, artificial neural networks yielded the highest F1 score of 72% for binary lameness detection. Furthermore, deep learning models were applied to identify different pressure distribution patterns with an F1 score of 88%. It was found that depth-first search was highly effective in locating and extracting information from isolated pressure areas. The kinematic measurements included twenty markers on different positions of 69 walking passages from twenty-three different cows, measured over three periods. Target variables comprised six gait attributes: leg swing, back arch, track up, flexion, asymmetric gait, and weight bearing, as well as an overall locomotion ability score. The prediction task was a multiclass classification of specific locomotion scores ranging from 1 to 5. A novel approach was proposed to analyze the trajectory of back markers as waves, resulting in an F1 score of 92% when combined with other gait attribute variables. The static kinetic measurement did not show satisfactory results regarding the prediction of locomotion score compared to kinematic markers. However, deep learning and depth-first search methods enabled the identification of hooves prone to future hoof issues. Furthermore, analyzing gait attributes from kinematic data – particularly the track-up and the trajectory of back markers – produced a satisfactory outcome and suggested the potential value of kinematic measurements as an early-lameness detection tool.