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

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%, respectively. In addition, the detection value was highly correlated with the manual measurement value of 25 pigs in the pigsty. The average relative error of body length is 2.56%. The average relative errors of height, chest width and abdominal circumference were 2.32%, 3.89% and 4.51%, respectively. In addition, in the farm, the accuracy of body size parameters is in accordance with the results of the laboratory. The experimental results show that the study is helpful to evaluate the body condition of pigs fed with concentrate and managed by breeders automatically and accurately.