Dog owners are highly motivated in understanding behavior and physiology of their pets and monitoring their wellbeing. Monitoring with a commercially available activity trackers reveals levels of daily activity and rest but recognizing the behavior of the dog would provide additional information, especially when the dog is not under supervision. In this study, a performance of a 3D accelerometer movement sensor placed on the dog collar was evaluated in classifying seven activities during semi-controlled test situation with 24 dogs. Various features were extracted from the acceleration time series signals. The performance of two classifiers was evaluated with two feature scenarios: using all computed features and the ones given by forward selection algorithm. The highest overall classification accuracy for the seven behaviors was 76%. The results are promising pro improving classification of specific behaviors by relatively simple algorithms.
|Conference||International Conference on Animal-Computer Interaction|
|Ajanjakso||1/01/18 → …|
- activity monitoring