Inertial Measurement Based Heart and Respiration Rate Estimation of Dogs During Sleep for Welfare Monitoring.

The earlier stages of training for a puppy to become a guide dog incorporates spending time away from guide dog schools with volunteer raisers providing puppies a safe environment to grow while they learn basic obedience and receive socialization opportunities. Our earlier work produced a smart collar system containing behavioral and environmental sensors to objectively track relevant parameters during this process. In this paper, we present a feasibility study to use this collar to track physiological signals relevant to sleep and resting state. We developed algorithms to extract heart rate and respiration rate from a sleeping dog using a 6-axis inertial measurement unit (IMU) integrated into a wearable collar form factor. We collected ten data sets of two minutes and used acceleration signals to estimate heart rate with an accuracy of 94.3%, precision of 84.1%, recall of 88.8%, and F1 score of 0.864. In addition, we were able to estimate the the respiratory rate at an accuracy of 94.3% on average. Further, we used the 6-axes to reconstruct an ECG signal with a deep neural network which improved prediction with a precision of 98.7%, and recall of 86.9% for an F1 score 0.925. The extraction of physiological features from non-invasive IMU signals collected from the neck region during sleep is a first step in sleep and resting state characterization in guide dog puppies. This, potentially, can be used to provide insight into the health and welfare of the guide dog puppies and the relation between sleep performance and puppy training and selection process.