Over the past decade, the Precision Livestock Farming (PLF) concept has taken a considerable place in the development of accurate methods for a better management of farm animals. The recent technological improvements allow the raising of numerous motion sensors such as accelerometers and GPS tracking. Several studies have shown the relevancy of these sensors to distinguish the animals’ behavior using various classification techniques such as neuronal networks or multivariate statistical approaches without considering the actual movements of the animal. For most of these researches, recordings are usually being taken at low frequency limiting the signal processing options.

Mass consumption mobile devices, such as iPhones, have nowadays the possibility to record accurately user movements with their inertial measurement unit (IMU) which regroups state-of-the-art 3-D accelerometers, 3-D rotational speed sensors, 3-D magnetometers, GPS and wifi positioning capabilities. As these sensors aim to detect human behavior, it was hypothesized that these mobile devices could also offer an efficient measurement unit for animal behavior research. The aim of this research is to investigate the capabilities of cow behaviors detection on pasture using the IMU of an iPhone 4S.

The mobile device was fitted on the neck of grazing cows and the signals from the IMU were recorded at a 100Hz frequency using Sensor Data (Wavefrontlabs), a dedicated software readily available on the application store. Cows movements were simultaneously video recorded during the experiments to distinguish grazing, ruminating and other behaviors. The set of data was split in two, one for calibration and the other for validation. A white-box approach detection algorithm was developed using the IMU signals in the time domain. It was based on criteria defined from the visual decomposition of the specific sets of movements of the head or jaws characteristic of each behavior. Statistical thresholds were determined for each criterion. The detection accuracies were obtained by comparing the discrimination algorithm results on the validation database with the observed behaviors.

Preliminary results based on the time-domain analysis showed accuracies ranging between 84% and 100% for the detection of grazing and ruminating behaviors demonstrating the performance of using the IMU of these mobile devices as compared to other data existing in the literature. As a high sampling frequency was used, a refined signal processing method will be tested in future works to improve the detection accuracies but also to provide more information on the link between animals’ behavioral changes and different pasture characteristics.