多视角深度相机的猪体三维点云重构及体尺测量  

Three dimensional point cloud reconstruction and body size measurement of pigs based on multi-view depth camera

作　　者：尹令[1,2] 蔡更元 田绪红 孙爱东[4] 石帅[1] 钟浩杰[1] 梁世豪 Yin Ling;Cai Gengyuan;Tian Xuhong;Sun Aidong;Shi Shuai;Zhong Haojie;Liang Shihao(College of Mathematics and Informatics,South China Agricultural University,Guangzhou 510642,China;National Engineering Research Center for Swine Breeding Industry,Guangzhou 510642,China;College of Animal Science,South China Agricultural University,Guangzhou 510642,China;Institute of Food Safety and Nutrition,Jiangsu Academy of Agricultural Sciences,Nanjing 210014,China)

机构地区：[1]华南农业大学数学与信息学院,广州510642 [2]国家生猪种业工程技术研究中心,广州510642 [3]华南农业大学动物科学学院,广州510642 [4]江苏省农业科学院农产品质量安全与营养研究所,南京210014

出　　处：《农业工程学报》2019年第23期201-208,共8页Transactions of the Chinese Society of Agricultural Engineering

基　　金：2019年度广东省重点领域研发计划“牲畜非接触式智能识别技术研究与示范”(2019B020219001);农业农村部农业物联网重点实验室开放基金课题(2018AIOT-08);江苏省农业自主创新基金资助项目“农业物联网关键设备研发及应用示范”(CX(16)1006)

摘　　要：对活体牲畜三维重构,数据采集方式、快速配准融合方法、表型体尺测量方法缺乏成熟有效的方案,导致目前活体牲畜的自动体尺测量技术难以在养殖场中推广应用。该文以猪为研究对象运用消费级深度相机KinectV2从正上方和左右两侧3个不同角度同步获取在采集通道中自由行走猪的局部点云。局部点云采用邻域曲率变化法去噪,并运用基于轮廓连贯性点云配准融合,最后采用多体尺数据精确估算技术测定包括体长、体高、胸宽、腹围等数据。该文分别对比实验室中模型猪由传输带以5种不同速率经过通道和养殖场内25头猪逐一经过通道,2种情况下采集数据进行各项体尺测算结果。其结果显示模型猪在传输带上以0、0.3、0.6、0.9和1.2 m/s等5种不同速率下测量体长、体高、胸宽、腹围值与实测值的平均相对误差分别为1.77%、1.36%、2.74%和2.17%。养殖环境下对25头猪同样4种体尺值与实测值的平均相对误差分别为2.56%,2.32%,3.89%和4.51%。试验结果发现养殖场活体猪测量最小误差可以达到实验室环境下的效果,但是最大相对误差变化较大,其原因在于养殖场中猪自由行走采集数据时行为姿态发生很大变化。Body size measurement is a major way to understand the key parameters of livestock for precision livestock farming(PLF)and effective management of large numbers of livestock.Manual measurement is one of the most commonly used methods to obtain the growth status of livestock.However,manual measurements can be time-consuming,costly,and sometimes harmful to animals and feeders.In addition,due to the lack of mature technology in effective data acquisition,robust registration and accurate estimation of multi-body parameters,non-contact measurement of live pigs is often a difficult task.Therefore,the application of automatic measurement technology of livestock and poultry body size parameters in actual breeding has great challenge.To solve these problems,a new 3D reconstruction and measurement system is proposed.Three consumer-grade depth cameras are set on the right,left and top of the data acquisition channel.When the pig passes the best shooting area of the channel,the camera synchronously obtains the point cloud data.Using filtering methods such as Gaussian curvature,outliers of three-dimensional images such as balustrade and other point clouds that do not belong to the pig contour are extracted from the original point cloud,and then the preprocessed point cloud in the three views is reconstructed based on the sample consistency(SAC),and then the pig body size parameters including body length are used.The body height,chest circumference and abdomen circumference are measured by the accurate estimation technology of body condition.In different experimental analysis,we compared 5 groups of body size measurement data at different speeds in the laboratory,and compared the body size measurement results of 25 pigs in the pig farm.In the laboratory,pig models were moved at 0,0.3,0.6,0.9 and 1.2m/s.The results show that the average relative error between the body length measurement and the manual measurement is 1.77%.The average relative errors of height,chest width and abdominal circumference were 1.36%,2.74% and 2.17%,re

关 键 词：图像处理 算法 模型 猪 三维重构 体尺测量 KinectV2 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]