Overview
You probably do not own a thermal camera, and even if you do you may not know how to use it. For many people, the first and only association with thermal imaging was the COVID-19 pandemic, which prompted widespread deployment of thermal cameras to screen for elevated body temperature from a distance. So the short answer to the headline question is currently no.
Why Adoption Could Change
That said, consider how digital cameras evolved. Decades ago digital cameras were niche and rarely used for surveillance; now most phones include multiple cameras. Two forces drove that change: technological breakthroughs and shifts in the drivers of demand. The same forces are affecting thermal imaging. Technological breakthroughs include the ability to build sensors using standard CMOS processes, enabling mass production, rapid miniaturization, low power, and steep cost reductions. At the same time, thermal imaging is no longer confined to military or border control; social and civic needs are beginning to drive development.
Temperature record of statues, trees, and a park fountain. (Source: Adam Saibeier)
The photo was captured with a multi-pixel thermal imaging sensor. Each detector is sensitive to long-wave infrared (LWIR) radiation emitted by objects. Each pixel encodes the surface temperature of the viewed object and is rendered in color: brighter colors indicate warmer areas, while darker blues and purples indicate cooler areas.
Thermal camera attachments for mobile devices are becoming more common and affordable, and some rugged phones even integrate thermal imaging modules. Thermal imagers are beginning to appear in smart home and IoT devices, so it will likely become commonplace to own a thermal camera. Artists have demonstrated what can happen when creativity is applied to enhanced thermal perception, and social media could show new forms of visual exploration of our relationship with the environment. Those examples are simple, but there are other reasons thermal cameras may appear sooner than many expect.
Thermal Imaging for Face Detection and Anti-Spoofing
Face detection and recognition are widely used for daily tasks such as authentication, login, and access control. Many methods have been designed to defeat these systems through presentation attacks, for example by showing printed photos, replaying videos, or using realistic 3D-printed masks. These presentation attacks, often called spoofing, are a serious problem because people favor the convenience of face-based authentication despite the risk of unauthorized access.
This problem has driven research into reliable face anti-spoofing (FAS). Beyond standard visual (RGB) sensing, other sensor modalities such as depth imaging and short-wave infrared (SWIR) imaging have been proposed. However, challenges including sensor cost and the lack of mature software algorithms or large multimodal face datasets have limited widespread deployment.
What these spoofing methods share is that, although they may be visually convincing, they do not present the inherent temperature signals of a living person. That makes thermal imaging (LWIR sensing) a promising complement to visual sensing for FAS. Thermal sensing can also improve face detection under challenging lighting and may make face recognition more robust. However, adding a thermal modality to FAS has not been comprehensively explored, and practical solutions for broad adoption still face technical challenges.
Low-Resolution, Low-Cost Thermal Sensors
Most FAS demonstrations that use thermal modalities rely on high-resolution, research-grade cameras that cost thousands of dollars. The question is whether relatively low-cost, mass-produced, low-resolution thermal imagers—such as 80 x 62 pixel arrays—can perform adequately so they can be integrated quickly into phones, ATMs, cars, doorbells, and elevators. The answer can be yes. One feasible approach is to perform face detection in the visual domain and liveness detection in the thermal domain.
A more comprehensive approach is to perform face detection in both RGB and thermal domains simultaneously. The demonstration below uses deep neural network (DNN) models, which outperform classical temperature-threshold and range-based methods, especially in challenging conditions with thermal background clutter and partial occlusion. As shown in the bottom row of the figure, landmark extraction was successfully completed in some cases. See bit.ly/3meLB7d for a demo of RGB and thermal face detection features.
Snapshot of real-time face detection performed by a DNN model operating on a radiation measurement stream at 15 frames per second from a second-generation 80 x 62 pixel thermal imager. The bottom row (center and right) shows a model that produced landmark extraction.
That leads to another question: can biometric features be extracted from a person's thermal "signature" and combined with features from the RGB domain to move from face detection to face identification? This is an active area of research and development. One challenge is persistence: features must be present across the population, distinct for each individual, and measurable across different physiological states. Another challenge is invariance: features must be detectable regardless of environmental conditions.
Public Health and Epidemic Monitoring
Digitalization of an increasing and more diverse range of assets and services is inevitable, layered on top of human needs for comfort, convenience, and safety. In smart home and smart city contexts, face identification and anti-spoofing technologies are becoming more common and important. The appearance of cost-effective thermal imaging sensors aimed at the mass market is an important milestone toward next-generation multispectral face identification systems.
In his book How to Prevent the Next Pandemic, Bill Gates called for global action to limit future pandemic impacts. Although current viral strains have not triggered the same response, and large-scale fever-screening devices have fallen out of everyday use, historical records show that fever has been a primary symptom across many human viral outbreaks—from past plagues to seasonal influenza, and including SARS, MERS, and novel coronaviruses. Two capabilities that could help manage and contain future outbreaks are: early automated detection of locations with unusually high rates of elevated body temperature, and the ability to monitor and visualize the spread of an outbreak.
When thermal imaging sensors become as unobtrusive and ubiquitous as visible-light sensors—embedded in household devices and personal gadgets rather than mounted on entryways or tripods—these monitoring capabilities can provide real-time feedback to individuals and health authorities, subject to appropriate governance. The key is that thermal imagers are in the hands of everyday users. Applications built around sensors and supported by deep learning models can detect abnormal body temperature with high accuracy. Then the application can prompt the user to manually submit the reading anonymously, or, with prior authorization, automatically upload the reading with loose localization to an authorized portal, as conceptually shown in the figure below.
Conceptual representation of early outbreak detection and monitoring using thermal imaging sensors embedded in personal mobile or desktop devices with AI capabilities to detect abnormal body temperature.
Outlook
We judge that mainstream adoption is not far off. CMOS-based thermal imaging sensors can now be built into imagers as small as 9 x 9 x 5 mm, comparable in size to visible-light sensors used in phones, and manufacturable at scales previously not possible. Costs are moving into ranges that allow integration into personal mobile and fixed computing devices. Functionally, these imagers open a wide range of possibilities, from social applications to enhanced face identification to global health monitoring.
