CMOS vs IR vs Thermal Sensors for rPPG: Performance Comparison

The selection of a sensor is one of the most critical decisions for hardware OEMs and IoT device makers integrating contactless vital signs monitoring. The choice directly impacts accuracy, reliability, cost, and the potential operating environments of the final product. For remote photoplethysmography (rPPG), the three primary contenders are standard CMOS sensors, infrared (IR) sensors, and thermal imagers. Understanding the fundamental differences in their operating principles is key to evaluating the complex landscape of cmos ir thermal sensor rppg performance and selecting the right hardware for a specific application, from automotive driver monitoring to clinical kiosks.

According to a comparative analysis of rPPG sensor modalities, the green channel of a standard CMOS sensor can achieve a Mean Absolute Error (MAE) for heart rate of around 5.9 beats per minute (bpm), while some infrared sensors have shown a higher MAE of 11.1 bpm under similar conditions. Thermal sensors, operating on a different principle, have demonstrated cardiac pulse accuracy between 88% and 90% in specific studies.

The core trade-offs in CMOS IR thermal sensor rPPG performance

The performance of an rPPG system is inextricably linked to its input: the video data stream from a camera sensor. Each sensor type captures different information from the human face to extract the blood volume pulse.

CMOS (RGB) sensors, the most common and cost-effective option, are used for visible-light rPPG. They work by detecting the subtle changes in skin color caused by the absorption of light by hemoglobin in the blood vessels. The green channel of the RGB spectrum is typically favored as it has the highest signal-to-noise ratio for this purpose. The primary challenge for CMOS-based rPPG is its sensitivity to ambient light conditions and motion artifacts.

Infrared (IR) sensors, typically operating in the near-infrared (NIR) spectrum (e.g., 850-940 nm), measure changes in the reflection of IR light from the skin. As blood volume fluctuates with the cardiac cycle, the amount of reflected IR light changes. This makes IR sensors particularly effective in low-light or nighttime conditions where visible light is absent, a critical requirement for applications like driver monitoring systems. However, the signal quality can be lower compared to CMOS in ideal lighting.

Thermal sensors, which operate in the Long-Wave Infrared (LWIR) spectrum (8-14 µm), do not measure reflected light. Instead, they detect the heat emitted from the skin. The blood perfusion in the superficial vessels of the face creates a distinct thermal signature. As blood flow changes, this signature fluctuates, allowing for the extraction of the heart rate. This method is completely independent of ambient lighting and can operate in total darkness. It is also less susceptible to skin tone variations but is generally the most expensive option and can be affected by ambient temperature changes.

Performance Comparison: CMOS vs. IR vs. Thermal

Feature	CMOS (Visible Light)	IR (Near-Infrared)	Thermal (LWIR)
Operating Principle	Detects changes in reflected visible light due to hemoglobin absorption.	Detects changes in reflected infrared light from subcutaneous tissue.	Detects heat variations on the skin surface from superficial blood flow.
Optimal Conditions	Bright, stable, frontal illumination.	Low-light to no-light environments.	Any lighting condition, including total darkness.
Low-Light Performance	Poor; requires dedicated illumination or advanced model tuning.	Excellent; designed for low-light and night vision applications.	Excellent; completely independent of visible or IR light.
Cost	Low; widely available and integrated into most devices.	Moderate; requires specific IR sensors and illuminators.	High; specialized sensors that are more expensive to manufacture.
Susceptibility to Motion	High; subject to significant motion artifacts that require software correction.	High; also subject to motion artifacts, similar to CMOS.	Moderate; less affected by minor head movements but still requires a stable view.
Primary Vitals	Heart Rate, Respiration Rate, SpO2 (experimentally).	Heart Rate, Respiration Rate.	Heart Rate, Respiration Rate, Skin Temperature.

Industry Applications

The choice of sensor is dictated by the product's intended use case and operating environment.

### automotive driver monitoring

In-cabin sensing for driver health and attention is a rapidly growing field, driven by safety regulations from bodies like Euro NCAP. The environment is challenging, with highly variable lighting and the need for nighttime operation.

• IR and Thermal sensors are the leading candidates. Their ability to function in complete darkness is a non-negotiable requirement for monitoring a driver at night.
• CMOS sensors are generally not viable as a primary sensor for this use case without significant, dedicated visible-light illumination, which would be distracting to the driver.

### iot and smart home devices

For consumer electronics like smart displays, smart mirrors, or fitness equipment, the priorities are often cost and ease of integration.

• CMOS sensors are the default choice. They are already present in many of these devices, making rPPG a software-based feature upgrade.
• The operating environment is typically a well-lit indoor space, which aligns with the optimal conditions for visible-light rPPG.

### clinical and kiosk applications

Telehealth kiosks, patient intake tablets, and other point-of-care devices require reliable measurements in a relatively controlled setting.

• CMOS sensors are highly effective here, as lighting can be standardized and patient motion can be minimized with clear instructions.
• Thermal sensors offer an interesting secondary benefit in these applications by providing a contactless skin temperature measurement, which can be useful for fever screening.

Current research and evidence

The academic and commercial R&D communities continue to push the boundaries of sensor performance. Deep learning models have significantly improved the robustness of CMOS-based rPPG, making them more resilient to lighting changes and motion. Research by McDuff et al. (2017) demonstrated early on the power of convolutional neural networks for this purpose. More recent studies focus on multi-modal approaches. A 2022 study published by Wang and colleagues explored the benefits of fusing data from both visible light and infrared sensors to improve accuracy across a wider range of conditions. Thermal rPPG research, such as that conducted by Gaulton and colleagues (2021), has validated its use for heart rate extraction, achieving accuracy comparable to contact-based photoplethysmography (PPG) in controlled settings. The consensus in the research community is that there is no single "best" sensor; performance is a function of the sensor, the software model, and the deployment environment.

The future of multi-sensor rPPG

The future of rPPG performance likely lies not in a single sensor "winner," but in sensor fusion. By combining the data streams from multiple sensor types, for example, a primary CMOS sensor for high-fidelity signal in good light and a secondary IR sensor for low-light conditions, hardware manufacturers can create far more robust and versatile systems. This multi-modal approach allows the system to adapt, using the best available data source at any given moment. This is computationally more demanding and requires sophisticated software to manage and interpret the data, but it represents the most promising path toward clinical-grade accuracy in real-world, uncontrolled environments.

Frequently asked questions

• Which sensor is "best" for rPPG?

• There is no single "best" sensor. The optimal choice depends entirely on the specific use case, operating environment, cost constraints, and desired performance characteristics. CMOS is best for cost-sensitive, well-lit applications, while IR and Thermal are necessary for low-light and nighttime operation.

• Can thermal cameras truly measure heart rate?

• Yes, thermal cameras measure heart rate by detecting the minute temperature fluctuations on the skin surface (especially around superficial arteries like the carotid or temporal arteries) caused by the pulsatile flow of blood.

• Why is the green light channel preferred for CMOS-based rPPG?

• The green wavelength of light is most strongly absorbed by hemoglobin in the blood compared to other wavelengths in the visible spectrum. This provides the highest contrast and signal-to-noise ratio for detecting the blood volume pulse in the skin.

The trade-offs between sensor cost, low-light capability, and signal quality create a complex decision matrix for any hardware team implementing contactless vitals. As the technology moves from the lab to production hardware, a deep understanding of the interplay between sensor physics and model performance is essential for success. Circadify is addressing this space by developing custom-trained rPPG models that are optimized for the specific characteristics of your chosen camera, whether it's a standard CMOS sensor, a night-vision IR camera, or a high-end thermal imager. To discuss your specific hardware and use case, contact our engineering team for a custom build inquiry at circadify.com/custom-builds.