Technology for Anxious Dogs: Monitoring, Wearables, and Cameras

Wearable activity monitors and accelerometers

Accelerometer-based activity monitors attach to a dog's collar or harness and log movement data. Researchers describe these devices as part of a category of smart sensing systems capable of remote data collection and analysis that could objectively track animal welfare states (Jukan et al., 2017; DOI: 10.1145/3041960).

Devices require independent validation before they can be used reliably. The FitBark 2, a commercially available dog accelerometer, had no published external validation before a 2021 study examined it (Colpoys and DeCock, 2021; PMCID: PMC7999242). That study (n=26) found the FitBark was a valid tool for tracking physical activity in off-leash contexts. On-leash accuracy was lower (Colpoys and DeCock, 2021; PMCID: PMC7999242).

Accelerometry can also detect welfare-relevant changes in activity. A study using accelerometry in dogs with chronic osteoarthritis pain found that differences in nighttime resting and daytime activity were detectable using functional linear modeling of the full diurnal trace. Traditional summary statistics of the same data failed to detect a treatment effect (Gruen et al., 2019; PMCID: PMC6775071). This shows that activity monitoring can reflect underlying welfare changes and that the analysis method matters.

Sensor placement and classification accuracy

Sensor placement affects how accurately wearable devices classify dog behavior. A validation study (n=45) testing sensors at two locations found a harness sensor on the back achieved up to 91% accuracy for certain dog activities. A collar-mounted sensor reached up to 75% accuracy at best (Kumpulainen et al., 2021; DOI: 10.1016/j.applanim.2021.105393). These are ceiling figures for specific behaviors. Average accuracy across all behaviors may be lower.

Adding gyroscope data improved classification accuracy by 0.7–2.6 percentage points, depending on classifier and sensor location (Kumpulainen et al., 2021; DOI: 10.1016/j.applanim.2021.105393). Static postures — lying, sitting, and standing — were the hardest to classify, particularly with collar sensors (Kumpulainen et al., 2021; DOI: 10.1016/j.applanim.2021.105393).

This matters for anxiety monitoring. Anxiety-related behaviors such as freezing, vigilance scanning, and tense resting are static. These are the postures that wearable sensors classify least reliably. Movement-based sensors detect locomotion better than the subtle stillness cues tied to fear.

A sensor-based system was used to detect behavioral changes in a small case series of dogs (n=24) receiving a nutraceutical for behavioral disturbances. The sensor tracked movement-related changes successfully but could not distinguish rest from sleep (Di Cerbo et al., 2017; PMCID: PMC5407696).

Computer-vision and video monitoring systems

Camera systems that apply computer-vision algorithms represent a different approach from wearable sensors. Instead of inferring behavior from movement data, video systems analyze visual frames to classify behavioral states directly.

Atif et al. (2023; PMCID: PMC10054391) described a prototype system combining a two-stream deep learning architecture with an LSTM model to recognize dog behaviors and summarize activity across a recording period. The system achieved an average F1 score of 0.955 for behavior recognition with processing speeds compatible with real-time use (Atif et al., 2023; PMCID: PMC10054391). The prototype demonstrated potential to help owners assess their dog's health and welfare. Formal clinical validation with anxiety-specific behavioral categories has not been established (Atif et al., 2023; PMCID: PMC10054391).

Detecting anxiety-specific states — panting when not hot, pacing, self-directed behaviors — requires validated ethograms and labeled training data for anxiety presentations. Published computer-vision monitoring research has focused on general health and welfare indicators. Anxiety-specific classification remains an open research gap.

Owner attitudes toward pet monitoring technology

Pet monitoring technology raises questions beyond technical accuracy. A qualitative study of 155 dog owners (Linden et al., 2022; DOI: 10.1109/tts.2022.3207991) found that owners see technology as having a role in supporting their dog care. Researchers described this as reflecting closely entangled daily routines between owners and dogs.

Owner attitudes were not uniformly positive. The study found two competing views across all activity types. A "dream scenario" described technology augmenting the owner's caregiving. A "nightmare scenario" described technology taking over the owner's role and weakening the human-dog bond (Linden et al., 2022; DOI: 10.1109/tts.2022.3207991). Owners held largely positive views toward technology for chore-like tasks such as cleanup. Views on technology for play, training, or feeding were more divided (Linden et al., 2022; DOI: 10.1109/tts.2022.3207991).

The researchers argued that the current trend in digital pet technology — focused on enabling remote interaction — aligns more with the nightmare than the dream scenario (Linden et al., 2022; DOI: 10.1109/tts.2022.3207991). Technologies framed as informing owner decisions appear more aligned with owner values than those positioned as replacing owner presence.

Evidence gaps and limitations

The canine wearable and monitoring literature is small relative to the commercial market. Several limitations matter when applying these devices in anxiety-management contexts:

Frequently asked questions