S Viazzi, ,  A Schlagater-Tello, C Bahr, C EB Romanini, , C Lokhorst, D Berckmans

n this study, a new computer vision technique to automatically detect lameness in dairy cows was evaluated. A 3D camera system was used to extract the back posture of the animals from top view perspective in a fully automatic way. Four parameters to describe the curvature of the back of the cows were used by a decision tree to classify lame and not lame cows. The experiment was conducted in a commercial Israeli dairy farm. The classification performance of the 3D algorithm was evaluated against the visual locomotion scores given by an expert veterinary. A dataset of 273 cows served to train the model and a dataset of 906 cows to validate it. The analysis led to a sensitivity of 67% and a specificity of 90% on a 2-point level scale (lame or not lame) on the validation dataset. These results show that the application of a 3D camera in dairy farming is feasible and can be used in order to develop a fully automatic lameness monitoring tool in dairy farming.