Signal processing and behaviour recognition in animal welfare monitoring system

The paper describes the evaluation of captured 3D-acceleration data from collars of dairy cows regarding the prediction of the state of health. It optimizes an existing algorithm on the subject of the distinction of lying and standing activities of dairy cows. It tests the accuracy of the classifier at different sampling frequencies, window sizes and window step-sizes. A parameter selection is discussed regarding the accuracy and minimizing use of resources, due to restricted processing power and energy resources on the targeted embedded system for the prediction algorithm.