基于三维点云与几何先验补偿的牛体体尺测定OA

[Objective]To address the problems of low efficiency in manual body measurement and the lack of standardized measurement data in the large-scale breeding of dual-purpose cattle,this study aimed to construct a high-throughput measurement system based on 3D point clouds.[Method]Taking 203 adult Simmental cattle in Xinjiang as subjects,a"top-view+side-view"multi-angle camera array integrating five cameras was deployed.A hardware synchronization network was constructed to eliminate motion artifacts caused by the cattle's natural walking.The algorithmic pipeline included multi-camera registration based on the iterative closest point(ICP)algorithm and voxel grid denoising.Furthermore,addressing the visual blind spots and point cloud loss in the hind leg region during natural walking,a dynamic occlusion compensation algorithm based on geometric priors was innovatively proposed.Using manual measurement as the standard,system performance was evaluated using Bland-Altman plots and linear regression models.[Result]A natural walking speed of 1.5 m/s,the frame acquisition time was approximately 0.06 s,and the core algorithm calculation took 1.79 s,improving comprehensive efficiency by about 60 times compared with manual methods.The system demonstrated high robustness for rigid traits such as height at withers and hip height,achieving measurement accuracies of 95.94%and 96.60%,respectively,with error distributions exhibiting good unimodal symmetry.For the challenging hind leg semi-girth,the compensation strategy based on geometric priors effectively overcame camera blind spots under dynamic conditions,achieving an accuracy of 89.05%with significant data consistency and no proportional bias.[Conclusion]This study constructed a non-contact body measurement system for cattle based on multi-view 3D point clouds,achieving automated collection at 1.85 s.The research confirmed that the system's measurement accuracy for height at withers,hip height,and other rigid traits exceeded 95%.Addressing the hind leg occlusion issue,the proposed geometric compensation algorithm effectively filled visual blind spots.This technology effectively resolved the difficulties of phenotype data collection in complex dynamic environments,providing standardized data support for the genetic breeding of cattle and the development of precision livestock farming.