The COVID-19 pandemic tested every assumption public health systems had about screening for infectious disease at scale. Airports installed thermal cameras. Office buildings pointed infrared sensors at doorways. Hospitals set up triage tents with temperature guns. And the uncomfortable finding, documented across dozens of studies, was that most of these measures caught far fewer infected people than anyone hoped.
That failure exposed something worth examining: the technology for contactless health screening existed before the pandemic, but it had never been stress-tested against a real outbreak. Now, with several years of data and a clearer picture of what worked and what didn't, researchers are taking a harder look at what camera-based vital sign monitoring can actually contribute to infectious disease detection and outbreak surveillance.
"A temperature asymmetry of 0.55°C or greater between the lacrimal caruncle and the forehead discriminated between COVID-19 positive and negative patients with 82% accuracy, questioning the widespread use of absolute forehead temperature measurement as a screening tool." — Martinez-Jimenez et al., European Journal of Clinical Investigation (2021)
The thermal screening problem
The most widely deployed contactless screening technology during COVID-19 was infrared thermography, either handheld infrared thermometers or fixed thermal cameras. The logic was simple: fever is a common symptom of respiratory infections, so catching people with elevated temperature should catch infected people.
In practice, the logic fell apart in multiple places. A study by Martinez-Jimenez et al. at Hospital Central in Mexico enrolled 80 participants with COVID-19 exposure and mild symptoms. Their infrared thermography analysis found that simple forehead temperature readings had limited diagnostic value. The more useful signal turned out to be temperature asymmetry between different facial regions, specifically the difference between the lacrimal caruncle (inner corner of the eye) and the forehead. Using a random forest machine learning model, this asymmetry discriminated between COVID-19 positive and negative patients with 82% accuracy at a 0.55°C cutoff.
That's a more sophisticated approach than what was actually deployed at most screening checkpoints. The vast majority of real-world thermal screening during the pandemic used single-point forehead measurements, which multiple studies found to be unreliable. The British Columbia Centre for Disease Control published an analysis concluding that skin-surface temperature scans are unreliable for COVID-19 detection, citing the gap between skin temperature and core body temperature as a fundamental limitation.
An observational study published in Nature Scientific Reports tested three different infrared thermography systems against digital oral thermometers in an Australian hospital during the pandemic. The thermal imaging cameras had RMSE values between 0.81°C and 0.97°C, with accuracy affected by outdoor temperature conditions ranging from 0 to 45°C. Environmental variability turned out to be a bigger confounder than most facility managers had anticipated.
Comparing screening technologies for infectious disease detection
| Screening method | What it detects | Contact required | Deployment scale | Sensitivity for infection | Specificity | Key limitation |
|---|---|---|---|---|---|---|
| Infrared thermal camera | Elevated skin temperature | None | Mass screening | Low to moderate | Moderate | Misses afebrile/pre-symptomatic cases |
| Handheld IR thermometer | Forehead temperature | Near-contact | Individual | Low to moderate | Moderate | Single-point measurement, operator variability |
| Thermal asymmetry analysis (AI) | Facial temperature patterns | None | Moderate | Moderate (82% in pilot) | Moderate | Requires specialized camera and algorithm |
| rPPG (RGB camera) | Heart rate, respiratory rate | None | Mass screening | Unknown for infection | Unknown | Cannot measure temperature directly |
| PCR testing | Viral genetic material | Yes (swab) | Individual | High (gold standard) | High | Slow turnaround, requires lab |
| Rapid antigen test | Viral proteins | Yes (swab) | Individual | Moderate | High | Lower sensitivity than PCR |
| Wearable biosensors | HR, temp, activity changes | Yes (worn) | Individual | Moderate (AUROC 0.85) | High | Requires device adoption |
Where rPPG fits into infection screening
Standard RGB cameras used in rPPG systems cannot measure skin temperature. That's worth stating plainly because it gets confused often. What rPPG does measure, heart rate and respiratory rate, can still provide useful signals for infection screening, just through a different mechanism than thermal detection.
Infections that cause fever typically also produce elevated heart rate (tachycardia) and changes in respiratory patterns. A person running a 39°C fever will usually also have a resting heart rate 10 to 20 beats per minute above their baseline. If you're continuously monitoring heart rate in a population (say, residents in a care facility or employees in a workplace), a sudden cluster of people with elevated resting heart rates could signal an outbreak before anyone reports symptoms.
This approach has a parallel in wearable device research. A study published in Sensors by researchers developing syndromic surveillance through wearable data analyzed physiological signals from 16,687 participants. Their fever detection model achieved an AUROC of 0.85 with a false positive rate of just 0.8% at 50% sensitivity. The principle, detecting infection through physiological changes rather than self-reported symptoms, applies whether the data comes from a wristband or a camera.
A 2025 review by Pirzada et al. at the University of St Andrews, published in Frontiers in Digital Health, surveyed 96 studies on rPPG health applications. The review confirmed strong evidence for rPPG-based heart rate and respiratory rate measurement, with accuracy comparable to contact methods in controlled settings. The authors noted rPPG's infection control advantage: as a non-contact method, it avoids pathogen transmission between patients and equipment, a practical benefit during outbreaks.
Hospital and care facility applications
In hospital settings, the infection screening value of contactless monitoring isn't primarily about catching patients who walk in already symptomatic. It's about detecting hospital-acquired infections early. A patient recovering from surgery whose heart rate gradually rises from 72 to 95 over 12 hours, with respiratory rate climbing from 16 to 22, may be developing a healthcare-associated infection before blood cultures or clinical assessment would catch it.
The same logic applies in long-term care facilities, which proved devastatingly vulnerable during COVID-19. Continuous camera-based monitoring of residents' resting vital signs could flag physiological changes associated with early infection across an entire facility simultaneously, without requiring staff to take manual vitals on every resident multiple times per day.
Airport and border screening
Airport thermal screening became the most visible public health intervention of the pandemic, and it had the most evidence of limited effectiveness. Wired reported on the gap between what thermal cameras could detect (elevated skin temperature) and what public health authorities needed to detect (infected travelers, most of whom were afebrile or pre-symptomatic).
An rPPG-based system wouldn't solve the fundamental problem with airport screening: most infected travelers don't have measurable physiological changes during their incubation period. But rPPG could detect a wider range of physiological signals than temperature alone. Someone with an active respiratory infection might present with an elevated resting heart rate and altered respiratory pattern even if their forehead reads 37°C on a thermal scanner.
Lessons from the pandemic response
The pandemic generated several findings that matter for future contactless screening design:
- Single-parameter screening fails. Checking temperature alone misses most infected individuals. Effective screening needs multiple signals: heart rate, respiratory rate, temperature, and ideally symptom questionnaires or behavioral indicators combined.
- Environmental conditions degrade accuracy badly. A thermal camera calibrated in a 22°C office gives different readings when someone walks in from 35°C outdoor heat or -10°C winter cold. Several studies documented RMSE increases of 0.5°C or more due to ambient temperature variation alone.
- Pre-symptomatic detection remains the hard problem. The majority of COVID-19 transmission came from people who either hadn't developed symptoms yet or never would. No external screening technology, contact or contactless, can reliably detect infection before physiological changes occur.
- Population-level monitoring has more promise than individual screening. Rather than trying to identify a single infected person at a checkpoint, monitoring aggregate physiological trends across a population may be more effective for outbreak detection.
Current research and evidence gaps
Clinical validation of rPPG for infection screening specifically is still sparse. Most published rPPG accuracy studies test heart rate and respiratory rate against reference devices in controlled settings, not against infection status. That's a different validation question, and it hasn't been answered yet.
The FDA-approved rPPG system studied by researchers at the Veterans Affairs healthcare system (published in JMIR Formative Research, 2024) demonstrated feasibility for contactless vital sign measurement during telehealth visits. The study by Haque et al. focused on usability and patient acceptance rather than infection detection, but the infrastructure it validated, capturing vital signs through a video call, could be adapted for remote infection screening in future outbreaks.
A clinical validation study published in PMC in 2025 confirmed that rPPG-enabled contactless pulse rate monitoring provided accurate results in cardiovascular disease patients, supporting the technology's reliability for continuous monitoring in clinical populations. Extending this validation to infection-specific vital sign patterns is a logical next step that several research groups are pursuing.
What comes next for outbreak preparedness
The next pandemic, whenever it arrives, will find public health infrastructure in a different position than 2020. The failures of simple thermal screening are documented. The potential of multi-parameter contactless monitoring is understood in principle but not yet validated at scale.
Several technical developments are converging that could change the calculus. Deep learning algorithms for rPPG signal extraction have improved substantially since 2020, with better motion artifact handling and performance across diverse skin tones. Smartphone cameras have gotten more capable. And the deployment infrastructure for video-based health monitoring has been tested, if imperfectly, in telehealth and clinical settings.
Circadify has developed camera-based vital sign measurement technology that captures heart rate and respiratory rate from standard video. The company is exploring applications in health screening and surveillance contexts, including potential roles in infectious disease preparedness. The ability to passively monitor physiological signals across a population, without specialized hardware or physical contact, addresses a gap that the pandemic made painfully visible.
What the field needs now is prospective validation: studies that deploy camera-based vital sign monitoring in real screening settings and measure its ability to detect early infection against confirmed diagnostic testing. Until those studies are done, the potential is clear but the proof is incomplete.
Frequently asked questions
Can a camera detect if someone has a fever?
Infrared thermal cameras can measure skin surface temperature from a distance, and some studies have reported moderate accuracy for fever detection. However, skin temperature does not always correlate with core body temperature, and environmental factors like ambient temperature, recent physical activity, and sun exposure can affect readings. Standard RGB cameras used in rPPG cannot directly measure temperature but can detect elevated heart rate and respiratory rate, which may accompany fever.
How accurate is thermal screening for infectious disease?
Accuracy varies widely depending on the technology and conditions. Martinez-Jimenez et al. found that infrared thermography could discriminate between COVID-19 positive and negative patients with 82% accuracy using facial temperature asymmetry patterns, but simple forehead temperature readings alone were far less reliable. The British Columbia Centre for Disease Control concluded that skin-surface temperature scans are unreliable as a standalone COVID-19 detection method.
What role could rPPG play in outbreak surveillance?
rPPG can continuously monitor heart rate and respiratory rate without physical contact, which could flag physiological changes associated with early infection before symptoms become obvious. This passive monitoring capability could be useful in airports, hospitals, or care facilities as a screening layer, though it would supplement rather than replace established diagnostic methods.
Is contactless fever screening effective at airports and public spaces?
Evidence is mixed. During the COVID-19 pandemic, thermal screening at airports detected some febrile travelers but missed many infected individuals who were asymptomatic or pre-symptomatic. Studies suggest thermal screening alone has limited sensitivity for catching infectious individuals in mass screening settings, though it may have deterrent value and can identify some symptomatic cases.
Related articles
- Contactless Heart Rate Monitoring — How camera-based heart rate measurement works and its clinical validation across different settings.
- Contactless Respiratory Rate Detection — Analysis of video-based respiratory monitoring accuracy and clinical applications.
- Camera-Based Vital Signs in Emergency Triage — How rPPG is being tested in emergency department screening workflows.