Importance of lameness in dairy cows
   Lameness is one of the main challenges in dairy production, with a significant impact on animal health, welfare, and productivity, directly affecting the profitability of the system. It is considered an indicator of pain and discomfort, compromising bovine mobility and quality of life. Early and accurate detection is essential to enable timely interventions, promote animal welfare, and increase productive efficiency on dairy farms.
General project description
   This project aimed to investigate lameness in dairy cows through three-dimensional (3D) kinematic analysis obtained from top-view videos and the use of machine learning techniques. A set of videos was collected using stereoscopic cameras positioned above the corridor where cows walked after milking, returning to the barn. Each video was independently scored by ten trained experts using the DairyNZ locomotion scoring (LS) system. The study focused on two main aspects: analyzing inter-rater agreement in locomotion scoring and developing computational models capable of predicting the degree of lameness from automatically extracted kinematic features. This work integrates animal welfare, data science, and precision livestock farming technologies.
Importance and use of the repository
   The video repository developed throughout this project represents an important contribution to the scientific community and future research in the fields of animal welfare and automated lameness detection. With three-dimensional recordings and locomotion scores validated by experts, the database supports the development and validation of new computer vision and artificial intelligence techniques and enables behavioral studies on bovine locomotion. It is a valuable and open resource to advance automated monitoring tools in dairy cattle farming.
Repository information
   The repository includes 430 paired videos (215 top-view and 215 side-view), organized according to the classifications assigned by the evaluators.
   To determine the final LS for each video, the mean of the ratings was used. Thus, the classification was based on predefined intervals: videos with means between 0 and 0.49 were classified as LS 0; between 0.5 and 1.49 as LS 1; between 1.5 and 2.49 as LS 2; and between 2.5 and 3.0 as LS 3.
   In addition to the videos, a spreadsheet is available containing the scores for each video as rated by each observer, along with the mean, standard deviation, and coefficient of variation.
   This allows users to assess the consistency and reliability of the classifications. The dataset includes 32 LS0 videos, 109 LS1 videos, 60 LS2 videos, and 14 LS3 videos. All videos are in 2D and MP4 format; however, the 3D top-view SVO files can be requested by email at victoriapdgaia@gmail.com.
Acknowledgements
   The authors acknowledge support from the Center for Comparative Studies in Health, Welfare and Sustainability (CECSBE) – FMVZ/USP and Group for Agricultural Digitalization, Ambience and Innovation (GDAAI) – FZEA/USP.
   The authors also express their gratitude to the locomotion evaluators that assessed the videos, to the farms and personnel for supplying their animals and locations for the recordings and to all the collaborators that contributed to this research.
   This study was financed, in part, by the São Paulo Research Foundation (FAPESP), Brazil -Process Number 2023/00338-3.
   This work was carried out with the support of the Coordination for the Improvement of Higher Education Personnel (CAPES), Brazil – Financing Code 001.