基于单视角点云镜像的猪只体尺测量方法

Pig Body Size Measurement Method Based on Mirror Reflection of Single Angle Point Cloud

为了克服人工家畜体尺测量耗时、应激大和工作强度大等问题,提出了一种基于单视角点云镜像的猪只体尺测量方法。首先使用单Xtion深度相机采集包含猪体的场景点云图像序列,并人工筛选出包含背部弯曲程度较小猪体的场景点云图像,然后基于随机采样一致性算法和聚类分割算法自动化分割目标猪体并对其进行姿态归一化,检测单视角猪体点云对称面,并利用对称面镜像获取完整猪体,最后利用自主研发的体尺测量软件测量猪只体尺。试验结果表明,利用该方法测量体长的平均相对误差为5.00%,臀宽测量的平均相对误差为7.40%,臀高测量的平均相对误差为5.74%。该方法为猪只体尺测量提供了切实可行的新途径。

In order to overcome the problems of manual livestock body size measurement which is labor intensive,time consuming and stressful to the animals,a livestock body size measurement method based on the mirror reflection of single angle point cloud was proposed. A single depth camera Xtion was utilized to capture the sequences of point clouds of the scene,including pig and point clouds were stored in the computer,and then the scene point cloud,including the pig that bending degree of its back was not large was selected manually. Subsequently,the ground of the scene was deleted by using random sample consensus,the remaining scene point cloud was segmented based on Euclidean clustering,and then all the clusters were sorted according to the distance from the center of cluster to the coordinate origin,and the smallest one was extracted as the target pig. Pose normalization of the extracted pig was obtained by using a PCA based approach. Then,the pig symmetry plane was detected from the single angle point cloud by using the prior information of pig after pose normalization. Using the symmetry plane,the point clouds data of whole body surface of pig was obtained. Finally,by using the software developed by the institution of the authors,body size measurement was performed manually. The results showed that the average relative errors were 5. 00% for body length measurement,7. 40% for hip width measurement and 5. 74% for hip height measurement. This method offered a new way to measure the livestock body size.