Determination of standing-time of dairy cows using 3D-accelerometer data from collars

Abstract

The paper describes the evaluation of captured 3D-acceleration data from collars of dairy cows regarding the prediction of the state of health. It focuses as a first step on the distinction of laying and standing activities and develops a classifier for a target system with restricted memory and CPU-resources. Therefore, a two-step classification algorithm is developed so that a deployment of resource-intensive task to a backend system is possible. A data reduction is considered to minimize data-transmissions and power consumption. The developed algorithm reaches data reduction on the embedded system to at least 2.6 % and an accuracy up to 90 % for the distinction of laying and standing activities.