SHUTTER-LESS TEMPERATURE-DEPENDENT CORRECTION FOR UNCOOLED THERMAL CAMERA UNDER FAST CHANGING FPA TEMPERATURE

Conventional temperature-dependent correction methods for uncooled cameras are not so valid for images under the condition of fast changing FPA temperature as usual, therefore, a shutter-less temperature-dependent correction method is proposed here to compensate for these errors and stabilize the camera’s response only related to the object surface temperature. Firstly, sequential images are divided into the following three categories according to the changing speed of FPA temperature: stable (0°C/min), relatively stable (<0.5°C/min), unstable (>0.5°C/min). Then all of the images are projected into the same level using a second order polynomial relation between FPA temperatures and gray values from stable images. Next, a third order polynomial relation between temporal differences of FPA temperatures and the above corrected images is implemented to eliminate the deviation caused by fast changing FPA temperature. Finally, radiometric calibration is applied to convert image gray values into object temperature values. Experiment results show that our method is more effective for fast changing FPA temperature data than FLIR GEV.


INTRODUCTION
Thanks to the fast development of microbolometer focal plane array (FPA), thermography using uncooled thermal cameras has achieved more and more attention in recent years (Niklaus et al. 2007).Without the requirement for stabilized temperature sensor, higher spatial resolution, smaller detector pitch as well as lower power consumption could be designed.This makes thermal imagery more cost-effective, lighter and smaller (Bhan et al. 2009).Therefore, uncooled cameras are widely used in forest fire protection (Ambrosia et al. 2003), building thermal leakage detection (Hoegner et al. 2016,Westfeld et al. 2015), * Corresponding author night vision (Liu et al. 2012), face recognition (Socolinsky et al. 2003), CO2 gas leakage from vegetation imagery (Johnson et al. 2012), water contamination monitoring (Lega et al. 2010).
In order to make any of the above applications viable, radiometric calibration which refers to forming a quantitative relation between image gray values and temperature or radiance of object has to be done accurately.However, the biggest drawback for uncooled thermal camera is that the camera output depends not only on the object radiance but also on the time-variant sensor temperature, which means the This contribution has been peer-reviewed.doi:10.5194/isprs-archives-XLII-1-W1-619-2017calibration parameters need to be updated almost continuously.Therefore, removing the image response from FPA temperature is important to measure the object temperature accurately.
One FPA temperature-dependent correction method is using one or more blackbodies to perform an online recalibration during measurement (Kruse et al. 2001).
Although this approach holds for most of the satellite applications for example cloud imagery system (Thurairajah et al. 2005), it is not resource-efficient and practicable for terrestrial applications because there is no room for a large, heavy and expensive blackbody.In place of blackbody, shutter is widely used as an Subsequently, on-the-line, all the acquired images are directly applied in object temperature retrieval by using the stored calibrated parameters as well as the temperature-dependent models.

Temperature-dependent Model Determination
During the experiment, our thermal camera (FLIR Ax65, whose parameters are shown in Table 1) is put inside a chamber with ambient temperature changing from 10°C to 35°C and then back, viewing a four element Peltier blackbody (shown in Figure 1) which are set to 11.6°C, 26.4°C, 67.6°C and 36.8°Cconstantly.
The maximum changing speed of FPA temperature is more than 1°C/min, which is much more than the condition of other papers (0.5°C/min in Nugent et al.

Application of the Method
The workflow to use the proposed method is shown in The time-related calibrated result and deviation between the calculated temperature and actual blackbody temperature are presented in Figure 6, which shows that our method is much better than FLIR GEV.
Our method has a low absolute mean error (0.23°C , 0.51°C , 0.08°C , 0.40°C) and a low absolute maximum error (1.45°C , 1.50°C , 0.57°C , 0.77°C) for all of the regions.. On the other side, the FLIR GEV has a larger absolute mean error (1.17°C , 0.92°C , 0.56°C , 0.74°C) and also a bigger absolute maximum error (4.81°C , 2.64°C , 1.49°C , 2.02°C).Note that the final temperature deviation includes a maximum ±0.15°CFurthermore, our method could still be improved in the future by analyzing the relatively large error existed at some of the beginning stages when fast changing FPA temperatures happen.In addition, whether this method is stable or not still needs more cameras to testify.
equivalent uniform temperature source to complete the radiometric calibration (Nugent et al. 2014).However, shutter-based compensation approaches have to close the shutter regularly during the measurement which leads to interruption and decrease of maximum acquisition rate, so they are not fit for seamless and real-time applications.An alternative method is to combine FPA temperature with the updating of correction parameters (Budzier et al. 2015).There are several temperature-dependent correction methods available, such as Kalman filter (Torres et al. 2003), piecewise Lagrange interpolation (Liang et al. 2017) and multivariate regression model using multiple temperature inside the camera (Tempelhahn et al. 2016).The main drawback is that none of them take fast changing FPA temperature into consideration in their experiments.Fast changing FPA temperature is primarily influenced by ambient temperature as well as self-heating when used outdoors with wind or abrupt weather changes.Then, the original relation (known from laboratory calibration) between the FPA temperatures and the image gray values is not valid anymore, which would lead to wrong measurement without proper correction.This paper proposes a novel shutter-less FPA temperature-dependent real-time correction method which is effective for fast changing FPA temperature data.The main idea of the method presented here is, firstly, all of the temperature-dependent models including stable temperature-dependent model, fast changing temperature-dependent model as well as radiometric calibration model are established by datadriven estimation for only one time off-the-line.
(1, 2, 3, 4).Then the changes of FPA temperature, housing temperature and four pixels' gray values via time are shown in Figure2.

Figure 2 .
Figure 2. FPA temperature, housing temperature and four pixels' gray values via time

Figure 3 .
Figure 3. Second-order polynomial parameters determination We take one of the stable FPA temperature as reference,

Figure
Figure 4. Third-order polynomial parameters determination

Figure 5
Figure 5(a).Firstly, the sequential images and FPA temperatures are saved at the same time.Next, the FPA temperature-based second-order polynomial model is used to project all of the sequential images into the reference level under the reference FPA temperature.Then, all of the above corrected images are modified again to remove the influence caused by fast changing FPA temperature using derivative PFA temperaturebased third-order polynomial correction.Finally, the normally stabilized response is radiometric calibrated by applying Planck model to get the final object surface temperature.