Methodology for Collecting Multimodal Data for Evaluating Volumetric Characteristics of Dairy Cattle

This paper presents the development and validation of a comprehensive methodology for collecting reference multimodal (RGB-D) data for contactless estimation of volumetric-weight characteristics of dairy cattle. A spatial configuration of a three-point measurement network based on Intel RealSense sensors (D435 series) is proposed and substantiated, designed with consideration of technological constraints in livestock facilities. The mathematical algorithm for projective transformation of two-dimensional pixel coordinates into metric three-dimensional space based on depth matrices is described. Optimal optical parameters were experimentally determined: object distance of 2.0-2.5 m at an optical axis height of 1.0-1.4 m. Validation of the methodology on a static full-scale model confirmed high instrumental accuracy: the average relative error of automated linear body measurements (height at withers, body depth, chest width) compared to manual measurements did not exceed 2.66%, which falls within zootechnical standards. The research results establish a standardized protocol for creating representative datasets necessary for developing machine vision systems within the paradigm of precision livestock farming.