Detector of walking state for cows through support vector machine based on accelerometers and gyroscopes installed on ear-tags and collars

摘要

The number of times that a cow performs a walking can be indicative of the estrus, presence of lesions and welfare degree. In order to identify the characteristic movements associated with walking, it is common the use of inertial devices attached to the animal's leg. The procedure of the device's installation entails risks to humans, animals and to the device. This paper aims to clarify if the inertial sensors installed in the animal's collar or even in an ear-tag can be used to estimate patterns associated with the animal's walking. Two devices with IMU's (inertial measurement unit) were fabricated in order to collect inertial variables from both points (ear and neck of the animal). The data used was unfiltered and filtered with linear and angular Kalman filters, then with each of this set of data, a support vector machines was trained. The best result achieved was with the equal error rate (EER) of 17.5% for both cases (ear-tag and collar) with data filtered with a linear Kalman filter. This error is similar to the error reached with the device installed in the animal's leg.