Adversarial Attacks on Thermal Infrared Sensing Systems

This direction is merged into another direction: color-driven thermal modulation. Its last version stays readable and citable here.

Version 1 of 113 papers19 conceptsfrozen 2026-10-04

Thermal infrared (TIR) sensors are often treated as a robust alternative to visible-light cameras because they operate in total darkness and are less affected by shadows. However, recent research reveals that these systems are vulnerable to a unique class of physical attacks that exploit the fundamental physics of heat transfer and emissivity.

By manipulating the "thermal delta"—the temperature difference between an object and its background—researchers have demonstrated that it is possible to hide people from surveillance, spoof autonomous vehicle sensors, and even bridge air-gapped networks.

The Physics of Thermal Deception

Visible-light cameras detect reflected photons, but thermal sensors detect emitted radiation. This radiation depends on two factors: the actual temperature of the object and its emissivity (how efficiently it sheds heat).

Most adversarial attacks on TIR systems work by disrupting the thermal signature that a machine learning model expects to see. For example, a pedestrian detector like YOLOv3 looks for a human-shaped "hot" blob against a "cool" background. Xingxing Wei and colleagues have shown that by using thermally insulating materials like aerogel, an attacker can create Physically Adversarial Infrared Patches. These patches don't just hide the person; they are optimized in shape and location to break the geometric features the detector relies on, achieving success rates over 80% across different angles and distances.

Passive vs. Active Thermal Perturbations

Research in this field generally splits into two categories: passive materials that block heat and active devices that generate it.

Passive attacks leverage the target's own heat. Xiaopei Zhu and Jianmin Li demonstrated this with Infrared Adversarial Car Stickers. By applying low-emissivity aluminum stickers to specific parts of a car, they can "project" a digital adversarial pattern into the thermal spectrum. Interestingly, these attacks are most effective when the car is running; the high heat of the engine increases the contrast between the car's surface and the low-emissivity stickers, making the adversarial signal stronger.

Active attacks use external heat sources to inject signals. Xiao Li showed that even small light bulbs can be used as physical perturbations to fool pedestrian detectors. Taking this further, Jian Lou developed a Physical Backdoor where a small, heat-modulating device acts as a trigger. When the device is turned on, the thermal detector is tricked into misclassifying a vehicle or making it "disappear" from the system's logic.

Thermal Channels for Data Exfiltration

Beyond fooling object detectors, the thermal spectrum can be used as a covert communication channel. Because computers have internal thermal sensors to prevent overheating, they can "hear" changes in room temperature.

Yisroel Mirsky and Mordechai Guri demonstrated this with HVACKer. By compromising a building's internet-connected air conditioning system, an attacker can pulse the room temperature to send binary commands to malware on an air-gapped computer. This "thermal modem" bypasses traditional network security by using the physical environment as the transmission medium.

The Challenge of Environmental Noise

The primary defense against thermal attacks is the environment itself. Thermal signatures are notoriously "noisy." Stefan Savage noted in Heat of the Moment that while thermal cameras can recover PINs from plastic keypads by detecting residual finger heat, this attack fails on metal keypads (which dissipate heat too quickly) or in direct sunlight (which overwhelms the signature).

Similarly, My Kieu found that thermal detectors perform differently during the day than at night. During the day, ambient heat from roads and buildings reduces the thermal contrast of human targets. For an adversarial attack to work reliably, it must be robust enough to survive "thermal washout"—the moment when the background temperature matches the target temperature, rendering both the object and the adversarial patch invisible.

Papers in this version

History & citationVersion 1 · 2026-10-04

Merged into another direction · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_adversarial_attacks_on_thermal_infrared__f906fc1e,
  title = {Adversarial Attacks on Thermal Infrared Sensing Systems},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/adversarial-attacks-on-thermal-infrared-sensing-systems/rdr_f906fc1e6cc84fe6b5235f8b8eac2181},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-19},
}

This exact version (version 1, frozen text):

@misc{lacuna_adversarial_attacks_on_thermal_infrared__f906fc1e_v1,
  title = {Adversarial Attacks on Thermal Infrared Sensing Systems},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/adversarial-attacks-on-thermal-infrared-sensing-systems/rdr_f906fc1e6cc84fe6b5235f8b8eac2181/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-19},
}