Speaker
Description
In infrared thermography, the measured surface temperature of a scene at a given time is typically presented as a thermogram. When the objective is to interpret temperature over time, analysis commonly requires visual comparison of multiple thermograms acquired sequentially. In practice, apparent temperature differences may arise not only from actual thermal changes but also from measurement uncertainty originating from the thermal imager, acquisition conditions, environmental variability, or target surface-related factors. If such uncertainty is not explicitly considered, both qualitative inspection and automated processing may overinterpret noise-level fluctuations as meaningful temperature dynamics of the observed surface.
In this work, we introduce the Transition Thermogram (TTG), a novel thermographic representation designed to integrate transient temperature behaviour and measurement uncertainty into a single two-dimensional image. TTGs are generated directly from thermograms acquired at two or three time points. The method is based on computing pixel-wise temperature differences between time points and evaluating these differences relative to uncertainty thresholds. The primary threshold reflects the expanded uncertainty of the thermal imager (0.3 °C). Optionally, an additional application-dependent uncertainty threshold can be introduced to account for further sources of variability; for example, in facial thermography this threshold may be 1.3 °C. Temperature variations that do not exceed the camera uncertainty are explicitly classified as no change, thereby suppressing fluctuations indistinguishable from measurement noise. When the second uncertainty threshold is applied, temperature changes are further categorised according to different levels of metrological significance without inflating the instrument uncertainty.
TTGs encode both the direction and level of temperature change through colour. Colour hue represents the direction or qualitative trend of the temperature transition, while colour saturation reflects the magnitude of change relative to the applied uncertainty thresholds. When computed from two time points, TTGs provide a compact map distinguishing stable regions from those exhibiting warming or cooling. When three time points are considered, the representation naturally extends to display more complex transient behaviours, such as monotonic heating or cooling, as well as changes in the direction of the temperature trend. In all cases, the visual encoding is explicitly governed by uncertainty-aware decision rules, ensuring that displayed transitions are metrologically defensible.
The proposed visualisation method requires only spatial alignment of the observed surface across thermograms and does not depend on specific acquisition protocols or hardware modifications. However, spatial misalignment, such as that caused by subject motion in facial thermography, can be mitigated by replacing pixel-wise temperature values with ROI-based metrics such as mean temperature. Consequently, TTGs are suitable for a wide range of infrared thermography applications concerned with temperature dynamics, as will be demonstrated in a practical application. By collapsing uncertainty-qualified temporal behaviour into a single image, TTGs enable rapid qualitative assessment of spatially heterogeneous transient thermal processes that may be difficult to interpret using conventional thermograms alone.
The uncertainty-aware nature of TTGs makes them particularly valuable for data-driven analysis and allows reliable data-based decisions. By explicitly suppressing temperature variations that fall within measurement uncertainty, TTGs provide more robust representations for feature extraction, statistical analysis, and machine learning workflows. Overall, the TTG offers a metrologically rigorous framework for visualising and analysing transient thermal behaviour, providing a solid foundation for both human interpretation and automated (AI-based) processing of thermographic time-series data.
