Researchers in Israel developed a system that analyzes ordinary smartphone footage, tracking 46 distinct points around a canine’s eyes, ears, nose, and mouth.
By examining how these facial landmarks shift, the software learned to differentiate between animals filmed before intensive assignments and those recorded afterward, according to ZME Science.
Dogs cannot speak when they reach physical exhaustion during high-stakes deployments. The primary application targets working dogs deployed for explosives-detection or search-and-rescue missions. These animals rely heavily on active sniffing to sample complex chemical environments, drawing air through their noses to locate target scents.
At the same time, dogs depend on panting to regulate internal body temperature through respiratory evaporation, creating a direct physiological conflict between cooling down and gathering scent information.
A 2003 study demonstrated that dogs searching for explosives after strenuous exercise panted more, sniffed less frequently, and performed worse at detecting their targets. A 2025 study tested 11 explosives-detection dogs and found that unexercised animals completed searches in an average of roughly 30 seconds. Following physical exertion, that completion time rose to 45 seconds.
Environmental conditions compound the physical strain on working animals. A 2024 study showed that working dogs required roughly 3.5 times the concentration of a target explosive odor to successfully detect it under extremely hot and humid conditions compared to milder environments. Continued exertion under such stress undermines both operational performance and animal welfare.
Anna Zamansky and her team at the University of Haifa studied four professionally trained working dogs, including two search-and-rescue animals and two detection dogs. Handlers recorded the subjects using standard mobile devices before work and immediately following 20 to 45 minutes of intensive searching, yielding 177 distinct clips.
The software tracked movements across 46 anatomical points on each animal’s face, utilizing machine-learning methods and neural networks to analyze movement sequences. For dogs the system had previously encountered, the software correctly classified about 78% of the recordings. Accuracy dropped to 69% when evaluating entirely unfamiliar animals.
“The results suggest that strenuous activity leaves subtle, measurable traces on a dog’s face,” ZME Science reported, though researchers have not yet confirmed whether the software measures exact fatigue levels.
Interpreting canine facial expressions remains challenging due to breed variations in ear position and muzzle anatomy. A 2026 study involving researchers from the same Haifa laboratory revealed that human observers spent substantially more visual attention evaluating the faces of short-muzzled dogs than dogs with typical muzzle shapes.
How many facial points does the AI track on a dog’s face?
The system tracks 46 anatomical landmarks located around the animal’s eyes, ears, nose, and lips using ordinary smartphone footage.
What was the recognition accuracy rate for unfamiliar dogs?
The system achieved a 69% accuracy rate when classifying recordings of dogs it had not previously encountered during training.
Where was the research published?
The findings were published in Nature Scientific Reports.
Keep reading