Automated monitoring of resting in dogs

Abstract Dogs may be deprived of adequate rest in certain environments such as noisy shelters or kennels. Accelerometers have been used to assess gross activity in dogs, but do not appear to distinguish simple inactivity from the prone, head-down recumbency that typifies sustained rest in the species. We tested the use of collar-mounted data loggers that record changes in tilt in 3 dimensions for monitoring head-down recumbency. Twelve dogs were studied using both video-recording and data loggers for a total of 36 h, with behaviour recorded every 10 s. From the video-recordings, dogs were coded as “resting” during a given minute if they were in head-down recumbency on all six observations in the minute, or “active” otherwise. With the criterion that rest was indicated when mean tilt change in a minute was ≤10°, the data loggers correctly identified active and resting minutes in 87 ± 2 per cent (mean ± SE) of minutes recorded. Thresholds of 8 and 12° were nearly as accurate whereas the other thresholds tested (2, 4, 6 and 14°) were less accurate. For two dogs the accuracy was lower (70 and 79%) because their unusual restless movements while recumbent often exceeded the 10° threshold. Receiver Operating Characteristic (ROC) analysis also showed the best performance (Sensitivity 0.88, Specificity 0.81) at a 10° threshold. We conclude that the tilt feature of the collar-mounted data loggers provides a reasonably accurate and efficient means of identifying head-down recumbency in domestic dogs but accuracy is reduced in dogs that are unusually restless while recumbent.