Transformer-based Dog Behavior Classification with Motion Sensors
A transformer encoder for real-time dog behavior classification from back-mounted motion sensor data.
Abstract
This research studies dog behavior classification using inertial motion sensors and a transformer-based neural architecture. The model uses self-attention to capture temporal dependencies across accelerometer and gyroscope signals collected from back-mounted sensors, focusing on seven behaviors such as walking, trotting, galloping, sniffing, lying, sitting, and standing. The proposed encoder-only transformer architecture adds a lightweight classifier head to produce behavior probabilities from 2-second sensor windows. On an open-access dataset of motion data from 45 dogs, the method reaches 98.5% accuracy, outperforming time-series DNN baselines while reducing model size, computational complexity, and latency for resource-constrained deployment.
Key Innovations
Transformer encoder for motion-sensor recognition
Applies self-attention to inertial sensor time series instead of relying only on recurrent or convolutional sequence models.
Seven-class back-sensor workflow
Uses 2-second, 200×6 motion windows from 45 subjects to classify key behaviors from a single back-mounted sensor.
Deployment-oriented accuracy and efficiency
Reaches 98.5% accuracy with reduced latency and computational complexity for real-time or constrained environments.