基于点云采集设备的奶牛体尺指标测量

为验证Xtion在奶牛体尺测量上应用的可行性,该文以提高现有体尺指标测量技术的精度、效率及自动化程度为目标,选用Xtion作为采集设备,石膏奶牛模型和真实奶牛作为试验对象,在实验室环境下,采用高精度三维扫描仪扫描奶牛模型作为对比点云数据,以不同距离下Xtion采集的数据作为测试点云数据,通过统计误差定量分析数据精度和密度随采集距离变化的规律,以确定合适的采集距离。养殖场环境下,在小于1.2 m采集距离条件下利用Xtion获取奶牛点云数据,采用Meshlab对点云数据进行可视化和交互测量,定性分析阳光、体表材质等因素对获取点云数据质量的影响,并将交互测量与人工测量结果进行对比分析。结果表明,在遮挡太阳光和采集距离大于0.6小于1.2 m条件下,平均误差小于±5 mm,相对误差小于10%,Xtion作为点云采集设备用于奶牛体尺测量是可行的。

Cow body measurement based on Xtion

Abstract: Since the numbers and capacities of dairy cattle farms are increasing with every passing day, computer aided studies for the management of these farms become more important and are widely used in daily life. In this study, we assessed the potential of Xtion as a point cloud data acquisition system for the body measurements of Holstein cows in the context of dairy cow conformation. For this purpose, we selected Xtion for point cloud data acquisition and chose a dairy cow model and a dairy cow as the experimental subjects. In an indoor environment, three dimensional laser scanners with high precision were employed to get point cloud data of the cow model as comparison data, and Xtion was used for acquiring point cloud data at different distances away from the cow model as test data. After aligning the comparison data and test data by using commercial point cloud processing software, statistical errors between comparison data and test data including positive maximum distance, negative maximum distance, positive average distance, negative average distance, standard deviation, and RMS error were calculated by the professional point cloud software in order to quantitatively analyze the precision and density of point cloud data from Xtion at different distances. The test results indicate that the errors almost increased linearly with increasing of the distance between Xtion and subject. The average errors were less than ±5mm on the condition that the distance from Xtion to the subject were between 0.6m and 1.2m. The average errors were less than ±10mm on the condition that the distance between Xtion and the subject were between 1.2m and 1.8m. Maximum errors were always less than ±20mm. The density of point clouds decreased exponentially with increasing of the distance from Xtion to the subject to such an extent that body measurements were impossible to get on account of missing the mark points on the body. These results imply that the appropriate distance between Xtion and a cow is between 0.6m and 1.2m. Under the dairy cow breeding environment, body measurements including chest width, thurl width, rear leg side view, front nipple length, rear udder width, and rear udder height of cows were first determined manually, by direct measurement. Then we used Xtion to acquire point cloud data at a distance of less than 1.2 meters from the cow. Visualization of the point cloud data and interactive measurement of the cow body based on point cloud data from Xtion were performed by Meshlab so as to qualitatively analyze the sunlight and surface materials' influence on the quality of point cloud data acquired from Xtion under dairy farm conditions. The results indicate that Xtion failed to give normal operation under the sunlight due to the presence of the IR rays in the sunlight. Point cloud acquisition from Xtion may have some small holes due to the material and appearance of a live cow body. The relative errors of comparing the body measurements obtained by Meshlab with the manual measurements were less than 10%, which indicates that Xtion is appropriate for body measurements of Holstein cows.