Automatic Determination of Body Condition Score of Dairy Cows from 3D Images

The increase in number, and decrease in cost for devices giving digital three-dimensional images have implied that the number of application areas where three-dimensional images can be used has increased lately. Body condition scoring (BCS), an indirect measure ranging from one to five of the level of subcutaneous fat in dairy cattle, is widely adopted for research and field assessment or for management purposes on farms. The feasibility of using statistical measures in three-dimensional images to automatically determine the body condition score was examined at DeLaval and Hamra Farm in Tumba, Sweden. This master’s thesis describes a method for processing threedimensional images of cows by automatic localisation of anatomical reference points. Points around the hook bones and following the spinal ridge were easy to identify. A number of parameters describing local feature statistical properties were examined to identify the ones that best describe the body condition score of dairy cattle. When the full data set testing a model combining the most significant histogram and Fourier parameters was used, 100% of predicted BCS were within 0.5 points of actual BCS and 79% were within 0.25 points. This research demonstrates the potential for using three-dimensional images for assessing BCS. The automatic method is particularly good at following BCS-fluctuations in time. Future efforts should explore ways to minimise the spread of the measured parameters and identify significant anatomical parameters that can be automatically extracted.