Dog's Behaviour Classification Based on Wearable Sensor Accelerometer Data

Abstract

Sensor-based behavioral detection and classification can improve dog health and welfare. Since continuous monitoring is required, an energy-efficient solution is needed. The number of logging axes, sampling rate, and selected features of accelerometer data not only have a significant impact on classification accuracy in activity recognition but also on the sensor's energy needs. Three models are designed for detecting dog's activities namely, a Random Forest classifier (RF), a Convolutional Neural Network (CNN) and a hybrid CNN, i.e. a CNN fused with statistical features that retain knowledge about the global time series form. The models are validated using an experimental dataset consisting of six different dogs performing in eight different activities i.e. lying, sitting, standing, walking, running, sprinting, eating and drinking. The results indicate that using neck and chest accelerometer data sampled at 10 Hz is sufficient for high overall classification accuracies (96.44%) for the three models. The hybrid CNN is capable of excellent performance, detecting nearly 97.87% of the behaviours at 10 Hz with a class accuracy of 80 % or higher.