Understanding canine emotions has long challenged researchers and owners alike. In humans, automated facial analysis is an established method in affective computing, but applying similar systems to dogs—whose faces vary dramatically across breeds—has proven difficult. In a pioneering study published in Scientific Reports, George Martvel and Stefanie Riemer developed and tested a fully automated geometric morphometrics-based pipeline for quantifying dogs’ facial expressions under real-world conditions.
The researchers analyzed owner-recorded videos of dogs during New Year’s Eve fireworks and compared them to control footage from quiet evenings. Using a validated system of 36 facial landmarks derived from the Dog Facial Action Coding System (DogFACS), the AI model tracked subtle changes in muscle movement associated with different emotional states. Despite challenges posed by variable lighting, angles, and breed morphology, the system successfully extracted reliable emotion-related patterns.
Key results indicated that backward-drawn ears were the most consistent indicator of fear during fireworks exposure, aligning with previous manually coded data. Dogs also showed increased mouth opening, which may reflect panting due to stress or arousal. The exclusion of the ear pinnae landmarks due to breed diversity did not reduce analytical accuracy, suggesting robust generalizability of the model across morphologically diverse dogs.
This study represents a significant advancement in automated emotion detection for companion animals. By eliminating the subjectivity and time constraints of manual coding, the approach paves the way for applications in behavioral science, welfare monitoring, and even real-time emotion assessment via smart devices. The authors emphasize that such tools could transform how humans understand and respond to dogs’ emotional well-being in natural environments.
Ultimately, Martvel and Riemer’s work bridges artificial intelligence and animal behavior research, providing a scientific foundation for AI-driven empathy toward dogs. As these methods evolve, they may help improve how humans interpret, support, and care for their canine companions during stressful experiences.
Source: Martvel, G., & Riemer, S. (2025). Automated analysis of emotional expressions in dogs based on geometric morphometrics. Scientific Reports, 15. Published September 2, 2025.
