A camera watches a person's face and hands, and from barely perceptible changes in skin temperature it understands: they are struggling right now. Body heat is like invisible ink that shows up under stress. Blood rushes to some areas and drains from others, creating a unique pattern for each person. The challenge is to find the spots and seconds in that thermal signature that betray anxiety.
The authors built a system called FABLE-Therm. It takes several pre-trained neural networks that look at different body parts and fuses their outputs early on, rather than at the end. This preserves details: which skin area warmed up, and when. It's like having multiple witnesses describe the same event — if each tells their part, the picture is fuller than if you just average their words. In tests with people with opioid addiction, the system achieved an accuracy of 0.938 on the AUROC scale (the closer to 1, the better).
But accuracy isn't everything. The researchers noticed the model performs worse on certain groups of people. They analyzed why: part of the gap can be closed by adding more data about those the system currently understands poorly. But there remains a gap due to people simply being different. It's like a thermometer that measures temperature more accurately for people of one body type. Understanding this difference helps build fairer systems that don't let down a specific individual.
Facial skin temperature can change by fractions of a degree during emotional arousal — sensitive thermal cameras can pick this up.
In the TV series 'Almost Human,' police androids read thermal signatures to understand human emotions.