How do my smart glasses know my breathing patterns while I'm hiking?

A hiker climbs a switchback, breath quickening with the grade, and the smart glasses resting on the bridge of the nose quietly log a respiration trend that tracks the effort. To the wearer it feels effortless, almost invisible. To the engineering team that shipped the device, that single number represents one of the hardest signal-processing problems in wearable sensing: extracting a slow, low-amplitude breathing waveform from a sensor that is bouncing, swinging, and shifting against the skin with every footfall. A reliable smart glasses breathing monitor is not a feature you switch on. It is the product of a sensing pipeline tuned to a specific optical path, mounting position, and motion profile, and the gap between a demo that works seated and a product that works on a trail is where most programs stall.

"Motion artifacts pose a significant challenge to accurate respiration rate estimation from wearable devices, and accelerometer data is increasingly used to correct photoplethysmography signals during dynamic activity.", findings summarized from Pimentel and colleagues' work on respiratory rate algorithms and the 2024 wearable monitoring literature.

Why a smart glasses breathing monitor is harder than it looks

Breathing does not produce its own bright optical signal the way a pulse does. Instead, respiration shows up as a faint modulation riding on top of the cardiac signal. Whether the glasses use a contact photoplethysmography (PPG) sensor in the nosepad or a camera-based remote PPG (rPPG) approach pointed at facial skin, the device is really measuring tiny changes in blood volume and skin reflectance. Three well-documented mechanisms encode breathing into that signal:

• Respiratory-induced amplitude variation (RIAV), where breathing changes the strength of each pulse beat.
• Respiratory-induced intensity variation (RIIV), a slow baseline drift in the signal tied to venous return and intrathoracic pressure.
• Respiratory-induced frequency variation (RIFV), the rhythmic speeding and slowing of heart rate across the breath cycle, also called respiratory sinus arrhythmia.

In a quiet room, an algorithm can isolate those modulations and report a respiration rate within a breath or two of a clinical reference. On a hike, the same modulations are buried under motion energy that is often orders of magnitude larger than the signal of interest. Footstep cadence, head turns, jaw movement from breathing through the mouth, and the glasses sliding microns against oily skin all inject noise directly into the optical band where respiration lives, roughly 0.1 to 0.5 Hz. The core engineering task for a smart glasses breathing monitor is separating physiology from locomotion when the two overlap in frequency.

How the signal actually gets recovered

Modern wearable pipelines do not rely on a single clean source. They fuse the optical signal with inertial data and apply learned models trained on the kind of motion the device will actually see. The table below contrasts the dominant approaches and where each one breaks down.

Approach	How breathing is extracted	Motion tolerance	Best fit for glasses
Single-channel PPG, classical filtering	Bandpass plus peak detection on RIIV or RIAV	Low; degrades fast above light activity	Seated or stationary use only
Nosepad contact PPG plus IMU fusion	Optical respiration modulation, accelerometer used to subtract motion	Moderate; handles walking, struggles with running	Strong for eyewear form factor
Camera-based rPPG plus facial motion	Skin reflectance modulation combined with facial landmark motion cues	Moderate to high with multimodal models	Glasses with outward or world-facing optics paired to companion devices
Deep learning multimodal (PPG + accel + gyro)	End-to-end network learns to reject correlated motion	High; trained directly on activity data	Production target for active-use eyewear

The clear trend across recent research is fusion. A 2024 study by Kontaxis and colleagues demonstrated a convolutional neural network using smartwatch PPG, accelerometer, and gyroscope signals together, reporting respiration estimates that outperformed conventional signal-processing baselines during movement. The same logic transfers directly to eyewear: the inertial measurement unit already present for head tracking becomes a motion reference that tells the model which parts of the optical signal to distrust.

Industry applications for hardware makers

Active and outdoor eyewear

For sport and outdoor glasses, respiration is a high-value metric because it indicates exertion and recovery more responsively than heart rate alone. But it is also where motion is worst. Manufacturers targeting this segment need models trained on hiking, running, and cycling motion signatures, not lab-bench data. Research on smart eyewear with nosepad PPG has shown that static respiration estimates can be reliable with low computational cost, while running and stair climbing remain open problems, exactly the conditions outdoor buyers care about most.

Everyday wear and wellness

Glasses worn all day generate long stretches of low-motion data, ideal for trend tracking such as breathing rate during focused work or sleep-adjacent rest. Here the engineering priority shifts from aggressive motion rejection to power efficiency and drift stability over hours of continuous capture.

Clinical-adjacent and accessibility devices

Eyewear aimed at respiratory wellness or accessibility needs tighter agreement with reference instruments and clear behavior when the signal is unusable. A camera-specific vitals model that knows when to suppress an unreliable reading is more valuable than one that always outputs a confident but wrong number.

Current research and evidence

The evidence base for breathing estimation during motion has matured quickly. A 2024 evaluation published in Biosensors examined PPG-based respiration monitoring during high-intensity interval training and used frequency-domain accelerometer data to correct motion artifacts, showing that fusion-based correction extends usable accuracy into genuinely strenuous activity. On the contactless side, researchers at Samsung Research have published on respiration rate estimation from remote PPG in the presence of non-voluntary artifacts, and on multimodal breathing estimation that combines facial motion with rPPG from an RGB camera, reinforcing that no single channel is sufficient under realistic conditions.

The wearable PPG roadmap led by Peter Charlton and collaborators in 2023 framed respiratory metrics as a major expansion area for optical wearables while cautioning that motion robustness and cross-device generalization remain the limiting factors. A recurring finding ties all of this together: a model tuned to one camera, one wavelength, or one mounting geometry rarely transfers cleanly to another. Sensor noise characteristics, the optical wavelength used in the nosepad, frame rate, and the precise skin contact angle all change how breathing is encoded. That is the practical case for camera-specific and sensor-specific model training rather than a generic algorithm dropped onto new hardware.

The future of smart glasses breathing monitoring

Several directions are converging. First, deeper sensor fusion will become standard, with the IMU treated as a first-class input to the respiration model rather than a post-hoc filter. Second, on-device inference will push respiration models into the milliwatt power budgets that all-day eyewear demands, favoring compact networks distilled for a specific chipset. Third, confidence-aware outputs will mature, so a smart glasses breathing monitor reports not just a number but a reliability estimate that downstream apps can act on. Finally, expect more wavelength experimentation. Comparisons of infrared, red, and green PPG at the nosepad suggest the optimal choice depends on the activity and skin contact, which again points toward models trained for the exact optical stack a manufacturer ships.

The throughline is specialization. The hardware that wins will pair purpose-built optics and inertial sensing with a model trained on the device's own data distribution, validated against the motion its users actually produce.

Frequently asked questions

Can smart glasses really measure breathing while I am moving?

Yes, within limits. Light activity such as walking is handled well by fusion models that combine the optical signal with accelerometer and gyroscope data. Vigorous motion like running narrows accuracy, which is why models trained on the target activity matter so much.

Do the glasses watch my chest or my face to detect breathing?

Neither directly in most designs. Breathing appears as a slow modulation on the blood-volume signal measured by a nosepad PPG sensor or by camera-based rPPG on facial skin. The device infers respiration from those modulations rather than imaging the chest.

Why does breathing accuracy vary between different glasses?

Because the sensor, wavelength, frame rate, and mounting position all change how the respiratory signal is encoded. A model tuned for one optical path does not automatically generalize to another, so accuracy depends heavily on hardware-specific training.

What can manufacturers do to improve respiration accuracy?

Collect data on the device under realistic motion, fuse optical and inertial signals, and train a model matched to the specific camera or sensor and use case rather than relying on a generic algorithm.

Building a smart glasses breathing monitor that holds up on a trail, not just on a bench, comes down to a model trained for your exact optics, sensor, and motion profile. Circadify is addressing this space directly with custom-trained rPPG and camera-specific vitals models built around your hardware. To scope a build for your eyewear platform, start a custom build inquiry at circadify.com/custom-builds.