摘要
针对奶牛的体重预估精度和自动化程度较低的问题,本文以荷斯坦奶牛为研究对象,提出1种基于奶牛点云构建三维模型进行体重预估的方法。首先,利用Kinect相机提取奶牛的俯视与侧视点云。其次,通过计算头颈部与背部脊线的点云拟合直线夹角,自动筛选出适宜三维重建的标准姿态点云。再次,利用缺失点云的相邻点云补全缺失区域后,提取配准特征点,利用特征点进行点云的配准。最后,基于背部脊线构建对称面实现点云镜像,完成奶牛点云三维重建,最终根据三维模型体积预估奶牛体重。通过对91头奶牛进行的体重预估实验分析,结果表明:体重预估绝对误差在-19.23~+20.04 kg之间,相对误差在-2.96%~+2.90%之间。本文方法实现了较精准的奶牛体重自动预估,可为奶牛精准养殖提供技术支持。
Aiming at the problems of low accuracy and automation of cow weight estimation,a method of using double view cow point cloud to build a three-dimensional model and predict the weight based on the model was proposed.Firstly,the Kinect camera was used to extract the point cloud of the cow in top and side view.Secondly,the point cloud of the head,neck and back ridge was calculated to fit the straight line angle and automatically filter out the standard posture point cloud suitable for 3D reconstruction.Thirdly,the missing areas were completed by using the adjacent point clouds of the missing point clouds,and then the alignment feature points were extracted and used for the alignment of the point clouds.Finally,the point cloud was mirrored by constructing a symmetrical surface based on the dorsal ridge line,completing the 3D reconstruction of the cow point cloud and finally predicting the cow's weight based on the 3D model volume.The results of the weight estimation experiments on 91 cows showed that the absolute error of weight estimation ranged from-19.23 to+20.04 kg and the relative error ranged from-2.96%to+2.90%.The method achieved a more accurate automatic estimation of cow weight and could provide technical support for accurate dairy farming.
作者
冯凡
王克俭
司永胜
FENG Fan;WANG Kejian;SI Yongsheng(College of Information Science and Technology/Key Laboratory of Agricultural Big Data of Hebei Province,Baoding 071001,China)
出处
《河北农业大学学报》
CAS
CSCD
北大核心
2024年第4期101-108,共8页
Journal of Hebei Agricultural University
基金
国家重点研发计划项目(2021YFD1300502)
河北农业大学精准畜牧学科群建设项目(1090064).
关键词
奶牛
体重预估
点云
三维重建
KINECT
cow
body weight
point cloud
three-dimensional reconstruction
kinect
