Hot or Not? Robust and Accurate Continuous Thermal Imaging on FLIR cameras Titti Malmivirta∗, Jonatan Hamberg∗, Eemil Lagerspetz∗, Xin Li∗, Ella Peltonenyz, Huber Flores∗ and Petteri Nurmi∗x ∗Department of Computer Science, University of Helsinki, Helsinki, Finland yInsight Centre for Data Analytics, University College Cork, Cork, Ireland zUniversity of Oulu, Oulu, Finland xLancaster University, Lancaster, United Kingdom titti.malmivirta@helsinki.fi, jonatan.hamberg@helsinki.fi,
[email protected].fi, xin.li@helsinki.fi, ella.peltonen@oulu.fi, huber.flores@helsinki.fi,
[email protected].fi Abstract—Wearable thermal imaging is emerging as a powerful Max temp and increasingly affordable sensing technology. Current thermal Mean temp 28 Min temp imaging solutions are mostly based on uncooled forward looking infrared (FLIR), which is susceptible to errors resulting from 27 warming of the camera and the device casing it. To mitigate 26 ) C ° ( these errors, a blackbody calibration technique where a shutter 25 e r u t a r whose thermal parameters are known is periodically used to e p 24 m e calibrate the measurements. This technique, however, is only T accurate when the shutter’s temperature remains constant over 23 time, which rarely is the case. In this paper, we contribute by 22 developing a novel deep learning based calibration technique that 21 0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 uses battery temperature measurements to learn a model that Time (s) allows adapting to changes in the internal thermal calibration (a) (b) parameters. Our method is particularly effective in continuous sensing where the device casing the camera is prone to heating.