Study On Sheep's Body Size Measurement Based On Cross-angle Computer Vision

This research focuses on the problem with common feeding craft in intensive sheep farming of our country,especially in western china,and aims to achieve real-time body measurement with contactless and automatic means during the process of sheep' growth.Sheep is of a high gregarious animal and tend to form groups among familiar individuals.In this paper,a new method of measuring the sheep's body size in a restricted space was proposed,which is to build a specific-structure device to limit the sheep in a certain space,and then to take images of the sheep in three directions(top view,left view,right view)under the natural light conditions with CCD cameras installed previously at the device.Vision image analysis was used to measure the nine body sizes classified three categories:lenth,width and height.Considering the sheep' joints,flexibility and multi-body postures,the improvement and the test on reliability for the detection method were carried out.A contactless and automatic sheep weight estimation system was developed.The main conclusions are followed:(1)A non-contact measuring system for sheep's conformation parameters based on cross-angle machine vision technology was presented,including individual identification,location restriction,weight,visual image acquisition and information management.The tested result in farm shows that this method can ease the livestock measuring workload greatly and overcome the limitations of manual measurement.(2)An automatic foreground area extraction algorithm based on simple linear iterative clustering(SLIC)SuperPixels and Fuzzy c-means(FCM)clustering was proposed,by which can effectively eliminate the occlusion from the fence;and obtain the high quality foreground image under harsh conditions.(3)Extraction algorithm of measuring points in side view has good adaptation,and can effectively extract multi-body sizes in different position and posture.The complementary parameters of left and right views can improve the accuracy of the measurement system.The average of several measurements can reduce the deviation from the actual value obtained by single measurement of the multi-postures body posture.The body parameters of 10 ewes with body weight of 64.5±8.30Kg and age of 18-36 months were measured,and compared with the manual measurement parameters.The maximum relative errors of withers height,back height,rump height,body length,chest depth,and foreleg height were 4.73%,2.58%,2.55%,2.50%,3.95%and 5.86%,respectively.The errors of withers height,back height,rump height;body length are chest depth are less than 5%,and the errors of back height,rump height and body length are less than 3%,indicated that the results of the measurement are correct.(4)A symmetric center extraction algorithm for flexible symmetric body image was proposed and applied in the top view of a multi-body posture of the sheep.The results show that the symmetric center line extracted by the algorithm in the data measurement area can well express the symmetric center of the flexible symmetric body.To average the several measurements can reduce the deviation from the actual value obtained by single measurement of the multi-postures body posture.The maximum relative errors of chest width,abdominal width and rump width were 3.8%,4.0%and 2.9%,respectively.(5)A sheep weight estimation model was studied,and the tested results show that the nonlinear model for body size and body weight is more accurate and with a better correlation between estimated weight and actual weight.When using single factor modeling,the power regression model is better,while using multi-factor modeling,the RBF network and SVM network are more effective.