In a pioneering 2013 study published in PLoS ONE, researchers Linda Gerencsér, Ádám Miklósi, and colleagues developed a wearable device capable of accurately identifying and classifying dog behavior in real time. Their system combines a tri-axial accelerometer, a tri-axial gyroscope, and a support vector machine (SVM) classifier to translate physical movement data into distinct behavioral categories.
The study involved 24 dogs (12 Belgian Malinois and 12 Labrador Retrievers) performing a set of predefined behaviors including lay, sit, stand, walk, trot, gallop, and canter. Each dog wore the inertial sensor while being video recorded, and the behaviors were labeled accordingly for supervised training of the classification algorithm.
Using SVM models, the system achieved an impressive over 90% accuracy when trained and tested on the same individual dog. When generalized across multiple individuals, the accuracy remained strong at over 80%, demonstrating the system’s robustness. The study shows that machine learning-based behavior tracking is not only feasible but highly effective in freely moving dogs without the need for external visual monitoring.
This automated approach offers significant potential for a wide range of applications—from ethological research and veterinary diagnostics to the development of smart collars and training tools. The authors suggest that the system could be adapted to other animal species, creating new avenues for behavior analysis in varied ecological and practical contexts.
Ultimately, the study lays the groundwork for scalable, technology-driven observation tools that can enhance both scientific understanding and animal welfare monitoring.
Source: Linda Gerencsér, G. Vásárhelyi, M. Nagy, T. Vicsek, Á. Miklósi. PLoS ONE, Volume 8, October 18, 2013.
