Published in the Journal of Computación y Sistemas, this study focuses on how computer vision can decode emotion-driven behaviors in domestic dogs. Given dogs’ exceptional social attunement to humans—such as following vocal commands, reading moods, and recognizing facial expressions—the authors argue that objective, automated emotion assessment is valuable for both welfare and performance-based contexts.
The team developed an artificial vision–based computational method trained on a video database in which dogs were presented with positive and negative stimuli designed to elicit four distinct emotional categories: aggressiveness, anxiety, fear, and neutral. A dataset of 1,067 images was used to train multiple models, with the best-performing system employing transfer learning on the MobileNet architecture.
The resulting model reached a test accuracy of 0.6917, indicating meaningful—though not perfect—classification performance. The authors note that the method can identify a dog’s emotional state at a specific moment, signaling strong potential for refinement and broader application.
They emphasize that this technology could improve selection, training, and task execution in working dogs by providing trainers and handlers with real-time or automated behavioral insights. Additionally, the approach supports more objective evaluation of stress-related states, helping safeguard canine welfare in environments such as search-and-rescue work, assistance tasks, or shelter care.
The authors conclude that while further methodological optimization is needed, the proposed system provides a baseline framework for expanding computational approaches to canine behavior research—potentially transforming how emotional states in dogs are interpreted and applied in practical fields.
Source: Chavez-Guerrero, V. O., Pérez Espinosa, H., Puga Nathal, M. E., & Reyes-Meza, V. (2022). Classification of Domestic Dogs Emotional Behavior Using Computer Vision. Journal of Computación y Sistemas. Published March 30, 2022.
