How Cameras Measure Blood Oxygen Without a Finger Clip

The integration of physiological sensing into everyday devices is shifting from dedicated clinical peripherals to ambient optical sensors. For hardware original equipment manufacturers (OEMs) and connected-device engineers, extracting heart rate from video via remote photoplethysmography (rPPG) is becoming standard practice. However, executing accurate camera blood oxygen measurement presents a significantly steeper physics and engineering challenge. Traditional pulse oximetry relies on transmission optics, driving light through a translucent appendage like a finger or an earlobe. Replicating this measurement using standard or specialized camera sensors requires relying entirely on reflection optics. The transition forces a heavy reliance on the camera sensor's ability to discern minute variations in specific light wavelengths bouncing off the epidermal layers. Understanding how these optical models estimate oxygen saturation is critical for hardware teams designing the next generation of embedded health monitoring AI.

"The principle for rPPG-based SpO2 evaluation is based on the ratio of AC/DC ratios between two distinct wavelengths, mirroring the underlying Beer-Lambert law mechanics of contact oximetry but applied to reflected ambient or structured light." , P. Pirzada, D. Harris-Birtill, and G. H. Doherty, University of St Andrews (2023)

The physics of camera blood oxygen measurement

The underlying mechanism of camera blood oxygen measurement relies on the differential light absorption properties of human blood. Oxygenated hemoglobin (HbO2) and deoxygenated hemoglobin (Hb) absorb light differently across the electromagnetic spectrum. In traditional pulse oximetry, a sensor emits red light (around 660 nm) and infrared light (around 940 nm). Deoxygenated blood absorbs more red light, while oxygenated blood absorbs more infrared light. By calculating the ratio of light transmitted through the tissue at these two wavelengths, the system can estimate the percentage of oxygenated hemoglobin.

Camera-based measurement, or rPPG, must accomplish the same mathematical ratio but via light reflected off the surface of the skin, typically the face. Because the camera sensor only captures the light that bounces back, the optical signal is exceedingly faint. The camera captures a video stream where the subtle color changes of the skin, imperceptible to the naked eye, correspond to the pulsatile blood volume changes with each cardiac cycle. This signal contains two components: the direct current (DC) component, representing the static absorption of tissues, bones, and non-pulsatile blood; and the alternating current (AC) component, representing the pulsatile blood volume.

To extract a usable saturation estimate, the system filters the rPPG signal from at least two distinct color channels. In a standard RGB camera, models typically utilize the red and blue or red and green channels to approximate the absorption differences. In multi-spectral or specialized IR cameras, the system can target the ideal red and near-infrared wavelengths directly. The algorithm then isolates the AC and DC components for both wavelengths, calculates the ratio of ratios (the AC/DC ratio of wavelength one divided by the AC/DC ratio of wavelength two), and maps this final value to an SpO2 percentage using an empirically derived calibration curve.

Hardware modalities: a comparison

The success of extracting an SpO2 signal is highly dependent on the optical hardware. Not all image sensors are capable of capturing the granular data required for accurate rPPG blood oxygen estimation.

Metric	Traditional Pulse Oximeter	Standard RGB Camera (rPPG)	Specialized IR/Red Camera (rPPG)
Light Path	Transmission (through tissue)	Reflection (bouncing off tissue)	Reflection (bouncing off tissue)
Wavelengths Used	Red (660nm) & IR (940nm)	Red, Green, Blue (broad spectrum)	Tuned Red & Near-Infrared
Signal-to-Noise Ratio	Very High	Low to Moderate (ambient light dependent)	High (controlled active illumination)
Hardware Dependency	Purpose-built LEDs and photodiodes	Consumer-grade CMOS sensors	Custom CMOS with specific bandpass filters
Primary Use Case	Clinical, direct contact monitoring	Software-based ambient tracking	Automotive sensing, medical kiosks

Camera sensor requirements for SpO2 extraction

Estimating blood oxygen saturation from a distance is unforgiving of sensor noise. While a simple heart rate extraction can survive a noisy sensor because it merely needs to track the frequency of the pulsatile waveform, SpO2 requires an accurate measurement of the waveform's absolute amplitude. If the camera hardware introduces noise that alters the relative amplitude of the AC component in the red versus the green or infrared channel, the resulting ratio calculation will be fundamentally incorrect.

Hardware teams must ensure the selected camera meets strict criteria:

• Color depth and dynamic range: The sensor must have a high bit depth (10-bit or 12-bit is preferred over standard 8-bit) to capture the micro-variations in pixel intensity. A higher dynamic range prevents the signal from clipping in bright conditions or disappearing into the noise floor in low light.
• Uncompressed raw output: Video compression algorithms (like H.264 or H.265) are designed to discard imperceptible visual data to save bandwidth. Unfortunately, the physiological rPPG signal resides precisely in the data that these codecs compress out. Access to raw, uncompressed video frames is often necessary for accurate amplitude extraction.
• Frame rate stability: While high frame rates are not strictly necessary for SpO2 (30 frames per second is generally sufficient to capture the cardiac pulse), the stability of that frame rate is critical. Variable frame rates introduce temporal jitter that corrupts the frequency domain analysis of the rPPG signal.
• Optimal signal-to-noise ratio (SNR): The sensor must be paired with high-quality optics and lenses that minimize optical aberrations. Larger pixel pitches on the CMOS sensor help collect more photons, naturally improving the baseline SNR before the signal reaches the software layer.
• Illumination control: Since reflection-based rPPG relies on ambient light for RGB cameras, fluctuating light environments will instantly disrupt the measurement. Dedicated systems mitigate this by using controlled active illumination, locking the exposure and gain settings of the camera.

Industry Applications

The ability to extract blood oxygen saturation without a dedicated peripheral opens new architectural possibilities for hardware deployments across multiple sectors.

Automotive in-cabin sensing

Automotive Tier-1 suppliers are integrating advanced driver monitoring systems (ADMS) that track far more than simple eye gaze. By using the near-infrared cameras already situated on the steering column or dashboard, manufacturers are exploring rPPG to monitor driver physiology. Hypoxia, a condition resulting from low blood oxygen levels, can cause drowsiness and impaired cognitive function. While challenging due to the dynamic lighting and vibration inherent in a moving vehicle, customized camera-specific vitals models filter out motion artifacts to extract reliable physiological baselines.

Medical kiosks and telehealth terminals

Point-of-care medical kiosks located in pharmacies, corporate offices, and remote clinics represent a primary deployment ground for contactless SpO2 monitoring. These environments offer a semi-controlled lighting scenario where users remain relatively still. Instead of maintaining and sanitizing peripheral hardware like finger clips or blood pressure cuffs, kiosk operators can use high-end embedded cameras to capture heart rate, respiration rate, and SpO2 simultaneously.

Smart glasses and wearables

The evolution of augmented reality (AR) and smart glasses is driving the miniaturization of optical sensors. Manufacturers are experimenting with positioning tiny cameras on the inward-facing frame of the glasses to capture the rPPG signal from the highly vascularized skin around the eyes and temples. Because the sensor is fixed relative to the user's face, motion artifacts are theoretically reduced, providing a cleaner continuous signal for embedded health monitoring AI.

Current research and evidence

The academic literature surrounding contactless SpO2 monitoring emphasizes both the theoretical viability and the practical engineering hurdles of the technology. A comprehensive 2023 review by researchers P. Pirzada, D. Harris-Birtill, and G. H. Doherty at the University of St Andrews highlighted that while rPPG has seen explosive growth, the majority of research has historically focused on heart rate extraction. The team noted that simultaneous and robust tracking of SpO2 via camera remains an active frontier, limited primarily by the physics of reflection optics and ambient light variability.

Further field validation is emerging in real-world scenarios. In a 2024 publication, researchers at Utrecht University, including Marnix Naber, evaluated remote heart rate and physiological imaging via online webcams, testing the boundaries of rPPG algorithms on uncalibrated consumer hardware. Their findings indicate that while generic algorithms struggle with the variable compression and noise profiles of standard webcams, strictly controlled sensor pipelines yield significantly better results. Additionally, an early 2023 pilot study by the Department of Pediatrics at Bielefeld University investigated the use of rPPG for monitoring vital signs, including SpO2, in pediatric populations. The researchers aimed to evaluate the feasibility of removing obtrusive wiring from pediatric wards, highlighting the growing clinical appetite for robust, camera-based measurement systems.

The future of camera-based blood oxygen

The trajectory of camera-based oxygen measurement is moving away from generalized software approaches toward highly specialized, hardware-coupled models. The current limitation of many commercial rPPG kits is the assumption that a single algorithmic model can accurately interpret the subtle optical signals from any generic CMOS sensor. Because SpO2 estimation relies on precise amplitude ratios, variations in a specific camera's spectral sensitivity curve, lens distortion, and digital signal processing pipeline will skew the SpO2 calculation.

Hardware OEMs are realizing that achieving reliable accuracy requires treating the rPPG model as an extension of the camera hardware itself. The future lies in custom-trained rPPG models optimized for the specific sensor, lens, and lighting configuration of the target device. By feeding the neural networks training data captured exclusively on the device's actual production hardware, engineers can calibrate the AI to natively account for the specific noise profile and color space of that sensor.

Frequently asked questions

How does a camera measure blood oxygen without touching the skin? A camera measures blood oxygen using remote photoplethysmography (rPPG). It records the micro-variations in skin color caused by blood pulsing through the microvascular bed. By comparing the amplitude of these color changes across different wavelengths of light (such as red and infrared), the system can calculate the ratio of oxygenated to deoxygenated hemoglobin.

Why is ambient lighting a problem for camera SpO2 measurement? Ambient lighting changes the spectrum of light reflecting off the skin. If environmental light fluctuates, it alters the intensity of the signal the camera receives. Because SpO2 estimation requires a precise calculation of light absorption ratios, variable ambient light introduces heavy noise that can corrupt the amplitude readings.

Do standard webcams work for blood oxygen estimation? While theoretical models can extract SpO2 from standard RGB webcams using the red and blue color channels, the accuracy is often limited by heavy video compression, automatic white balance adjustments, and low signal-to-noise ratios. Production deployments typically require specialized sensors or uncompressed video feeds to achieve usable data.

What is the difference between extracting heart rate and SpO2 from a camera? Extracting heart rate only requires identifying the frequency of the pulsatile signal (how often the heart beats). Extracting SpO2 requires measuring the exact amplitude (the height) of the pulsatile signal across different color channels to calculate an absorption ratio. Amplitude measurement is far more susceptible to camera noise and lighting artifacts than frequency tracking.

For hardware OEMs and connected device engineers, transitioning from a conceptual contactless vitals demo to a production-ready feature requires moving beyond generic algorithms. tryvitalsapp.com builds custom-trained rPPG models optimized specifically for your chosen camera, sensor suite, and deployment environment. By aligning the embedded health monitoring AI directly with your hardware's unique optical characteristics, our team helps eliminate the friction of off-the-shelf software limitations. To explore architectural requirements or begin scoping a custom vital signs model for your next hardware lifecycle, submit a custom build inquiry today at circadify.com/custom-builds.