Abstract: Animal husbandry has been a major contributor to the economy of many countries, including the US and Canada. The integration of Artificial Intelligence (AI) has revolutionized this sector, mirroring its transformative impact across industries by enabling scalability through advanced technologies. In particular, AI has introduced automated monitoring and detection tools leveraging computer vision, which helps automate many of the tasks performed previously by human labor. One of the specific assets from the computer vision toolbox is pose estimation; an approach that enables precise identification and tracking of animal behaviors, akin to human observation processes. This innovation addresses the challenge of scalability in cattle management by confronting health complexities and welfare concerns. Our study reviews pose estimation methodologies and their applicability in improving animal husbandry. Furthermore, we propose an initiative to enhance open science within this field of study, by launching a platform designed to connect industry and academia, with the aim of tackling existing challenges and promoting advancements in the field.

Abstract: Bovine motor disorders can be difficult to diagnose in their initial stage. Quantitative gait analysis methods have been designed to solve this problem, but the on-site subjective assessment of the gait pattern remains prominent and depends on human expertise. Gait assessment in feedlot cattle can be done using Artificial Intelligence (AI) as a substitute for the absence of human diagnosis, but creating an AI diagnosis procedure requires substantial behavioral information for training the AI tool. One solution for obtaining behavioral information is to use AI-assisted tools for diagnosis based on recordings of cattle movement. In this study, we created a three-dimensional digital representation of walking cattle to generate the required information and compare its applicability to that of the actual gait patterns. We used video recordings of cattle walking and trotting, and then used them as reference to create three-dimensional pose representations. Then, we introduced variations to these representations by altering specific aspects of the original walking cow model and its environment. We then tested the combined representations against the real data to see if they can increase the efficiency of the real for training a deep neural network for detecting gait analysis. This method can compensate for the scarcity of behavioral data, and enable researchers to amend shortcomings of their data.