Influence of image resolution on the performance of remote breathing rate measurement using thermal imaging technique

Abstract

The aim of this work was to evaluate the influence of a thermal image resolution on the performance of thermal imaging system for remote and unobtrusive breathing rate (BR) measurement. The thermal imager with original resolution of 640 × 480 was first used to establish the thermal imaging system. The thermal video database containing 60 videos was developed to validate the feasibility of this system for BR measurement. Subsequently, the nearest-neighbor interpolation was exerted to create the thermal videos with low resolutions. The Bland-Altman plots of BR showed that distances between the upper and lower 95% limits of agreements were 4.094 bpm, 10.595 bpm, 10.091 bpm, 15.168 bpm and 20.769 bpm for video resolutions of 640 × 480, 512 × 384, 320 × 240, 192 × 144 and 64 × 48, respectively. With respect to linear regression analysis, the determination coefficient (R2) of 0.898, 0.890, 0.900, 0.776 and 0.539 had been found for the resolutions of 640 × 480, 512 × 384, 320 × 240, 192 × 144 and 64 × 48, respectively. These two statistical approaches both demonstrated that the thermal resolution of 320 × 240 achieved the performance comparable to the high thermal resolution. Inspired by the above results, for the practical applications, we can choose the appropriate resolution by taking both the system performance and cost into account. In this case, the thermal video resolution of 320 × 240 could be selected for BR measurement in some application scenarios (i.e. lie detection) where the measurement distance is approximately 1.5 m. This research may greatly improve the design of the contactless measurement device for physiological signals.