Published in Advanced Robotics, this study by Miiamaaria V. Kujala, Anna Valldeoriola Cardó, and ten colleagues assessed how well three devices—ActiGraph GT9X Link, Kaunila, and FitBark—classified canine behaviors based on acceleration patterns.
The sample included 70 adult companion dogs who performed seven behaviors: three still behaviors (lying, sitting, standing) and four dynamic behaviors (walking, sniffing, trotting, playing). Tasks were repeated across two sessions, and all behaviors were confirmed via video for ground truth accuracy.
All devices produced correlated activity scores, and median acceleration values reliably separated still from dynamic behaviors. Dynamic behaviors could also be distinguished from one another, except for walking vs. sniffing when using the back-mounted ActiGraph GT9X or the Kaunila collar-mounted device—highlighting a shared movement pattern during these activities.
Accuracy of behavioral cutoffs ranged from moderate to high, with standing and walking being the most difficult behaviors to classify precisely. Overall, the cutoff-based classification accuracy was approximately 80% for all devices, demonstrating that commercial trackers perform reasonably well in practical use.
Importantly, device output was not affected by dog age, weight, or sex, suggesting that these trackers offer consistent results across diverse canine populations.
The study concludes that while commercial activity trackers are not perfect, they are reliable tools for monitoring canine activity and distinguishing between many everyday behaviors—making them valuable for pet owners, veterinarians, and researchers.
Source: Kujala, M. V., Valldeoriola Cardó, A., Vehkaoja, A., et al. (2024). Comparison of activity trackers in estimating canine behaviors. Advanced Robotics. Published April 28, 2024. Research conducted across interdisciplinary teams in canine cognition, sensor technology, and robotics.
