10 th International Conerence on Quantitative InraRed Thermography July 27-30, 2010, Québec (Canada) New method or two-point nonuniormity correction o microbolometer detectors by R. Olbrycht*, B. Wiecek*, T. Swiatczak* *Electronic Circuits and Thermography Division, Technical University o Lodz, Institute o Electronics, Poland, robert.olbrycht@p.lodz.pl, wiecek@p.lodz.pl 1. Introduction In this paper, the new method o two-point nonuniormity correction o microbolometer detectors is presented. Microbolometer detectors are commonly used in thermal cameras thanks to small dimensions, relatively low price and no requirement or cooling. However, there is a common problem o detector non-uniormity, due to very high thermal sensitivity o the microbolometer structure. Thermograms obtained directly rom a detector, without any correction, are almost unusable. The inluence o microbolometer detector housing temperature variations results in the appearance o alse temperature gradients in thermograms. This phenomenon causes improper temperature measurement. An example o above mentioned temperature gradient (visible in the background) is shown in Fig. 1 (b), while the corrected thermogram is shown in Fig. 1 (a). Markers in those thermograms display the temperature values, in C. One may notice that the temperature readout error prior to correction peaks at about 16,6 C in this particular example. a) b) Fig. 1. a) corrected thermogram, b) thermogram prior to nonuniormity correction 2. Nonuniormity correction traditional one-point approach Under the assumption that the characteristic o a single microbolometer can be described by a linear equation (1), there are two parameters that can be corrected or each microbolometer oset and gain. [1] ) gi, + oi, = φ (1) where: Y(T i, ) is the response o a microbolometer on position i, in a detector to an IR radiation lux Φ corresponding to the temperature T i,, g i, is the gain o this microbolometer, o i, is the oset. It is essential to provide the compensation or detector nonuniormity during the measurements carried out with a thermal camera. The commonly used approach relies on the mechanical shutter, which is periodically introduced in ront o the detector surace. This method is called one-point NUC (NonUniormity Correction), because it can correct only the oset
parameter (o i, ). The drawback o this method is that every minute shutter disturbs the thermal observation resulting in periodically rozen image or one to two seconds. It is assumed that all microbolometers have gain values corrected at the actory, and there is no need to repeat this correction. However, in [2], it was shown that in practice the gain parameter also changes with time, partly because o the ageing o microbolometers, but also or example due to the ubiquitous dust that may be present on the surace o inrared lens or detector. 3. Nonuniormity correction the proposed two point method The proposed method combines both the oset and gain nonuniormity correction. Its principle is based on using a removable inrared ilter or oset correction, and an inrared emitter or gain correction. The ilter has to be semi-transparent or inrared radiation. The IR emitter is an additional source o inrared radiation, which stays out-o-ocus and provides the excitation to the detector in parallel with the radiation o observed scene. The main advantage o proposed method is the lack o interruptions o thermal observation during the correction. The scheme o the measurement rig is shown below. a) b) Thermal camera Inrared emitter (c) Observed scene c) Uniorm surace Removable inrared ilter (b) Fig. 2. The scheme o the measurement rig (a), with thermal camera looking at the scene through the inrared ilter (b) and subected to the inrared radiation rom the emitter (c). 3.1 Gain non-uniormity The proposed gain non-uniormity correction approach is based on using an inrared emitter. Its purpose is to subect the microbolometer matrix inside the camera to an additional inrared radiation in parallel with the radiation coming rom the observed scene. In case o experiments presented in this paper, this element is localized in ront o the lens, close to its surace, hence it remains invisible or the camera until turned on (stays out o ocus), as shown in Fig. 2. In the inal applications, however, it is planned to localize it inside the camera, but not in the optical path o detector-lens-scene. The emitter radiation may be modulated with the sinusoidal or square unction with requency o about 20 to 25 Hz. This is the highest ramerate that typical microbolometer camera can record, while the observed scenes rarely contain such requency components. The response o the detector solely to the IR emitter excitation may be calculated using requency analysis (e.g. FFT), and it contains the inormation about the gain non-uniormity o this detector. This method enables a thermal camera to perorm a live gain correction without using a shutter. It is possible to use a simpler approach, which assumes capturing two consecutive rames in the shortest possible time period the irst one with the additional radiation, the second one without. The dierence between those rames contains the gain non-uniormity inormation. In practice, in both cases it is required to provide the compensation or the uneven illumination o the microbolometer matrix by the emitter. The requency component o the emitter may be eventually removed rom the thermal image delivered to the camera screen. The detailed description o the proposed method or gain-non uniormity correction is provided in the patent application [3].
3.2 Oset non-uniormity The proposed method or the oset non-uniormity correction relies on a removable inrared ilter, which is periodically introduced in between the camera lens and observed scene (similarly as a shutter in traditional NUC solutions is inserted between the lens and detector). Authors ound out that inclining the ilter protects against lens relections in the ilter surace. Knowing the ilter transmittance or inrared radiation, it is possible to calculate the microbolometer matrix oset map, and use it to perorm oset correction, taking into account the previously calculated gain nonuniormity inormation. Thereore the proposed method is a two-point correction, and it is believed to be more accurate than the one-point approach. During the instantaneous ilter usage it is possible to compensate or its transmission to deliver the live thermal image to the camera screen. The ilter with an uniorm transmittance actor τ is an additional source o inrared radiation with emissivity ε, according to the Kirchho s law: ε = 1 τ ρ. (2) It is required to acquire two thermograms (with the shortest possible time delay) beore introduction o inrared ilter, and directly ater inserting it in ront o the microbolometer detector. Between those acquisitions the camera and scene is assumed to be stationary. Beore ilter introduction the readout o microbolometers may be described by the equation (1), assuming the linear character o response to a narrow range o scene temperatures. In case o the second thermogram (with the ilter), the response Y* o microbolometers may be described by the ollowing equation: where: [ φ( Ti ) + φ( T ) (1 )] oi * ) = gi,, τ +, τ (3) Y*(T i, ) is the response o a microbolometer on position i in a detector to an IR radiation lux Φ corresponding to the temperature T i,, ater introducing the ilter, τ is the transmittance o this ilter, φ(t ) stands or the IR radiation lux corresponding to the ilter temperature. The solution o (1) and (3) or φ(t i, ) and o i, is given by (4) and (5), respectively [4]: * φ Yi Y i, ( Ti ) φ( T g (1 τ ) + ), = (4) i, Y y τ * i, i, o i = φ( T ) gi, 1 τ (5) The lux o scene radiation, irrespective o the microbolometer temperature drit, is given by (4). Applying (4) or all microbolometers in the matrix, one can perorm the two-point non uniormity correction. In result, the real thermogram o observed scene may be obtained, without measurement errors introduced by the temperature drit phenomenon, that may be calculated using (5). 4. Correction results Fig. 3 (a) presents the reerence thermogram, which is not burdened with any measurement errors. I the nonuniormity correction is abandoned, the alse temperature gradients appear due to the thermal drit phenomenon. With time, the measurement error arises, what is shown in Fig. 3 (b) scene temperatures remained unchanged, but microbolometer readouts are dierent than in case o Fig. 3 (a). Let s suppose that at this point one wants to perorm the two point nonuniormity method using the proposed approach. The irst required thermogram is Fig. 3 (b), as seen by the camera. Instantaneously one should introduce the inrared ilter (as shown in Fig. 2) and record the second thermogram - Fig. 3 (d). Ater the ilter is removed, the inrared emitter is turned on, and third thermogram is acquired - Fig. 3 (c). It is important to keep the lowest possible time delay between recording those thermograms to ensure that scene content and camera position does not change in between. Alternatively it is possible to extract thermogram Fig. 3 (c) using requency analysis, when inrared emitter pulsates with a given requency. In this case the camera is not required to be stationary.
Fig. 3. Input thermograms: a) reerence thermogram not burdened with the temperature drit, b) thermogram (a) burdened with the temperature drit, c) thermogram (b) with additional radiation rom IR emitter, d) thermogram (b) recorded through the inrared ilter Having these thermograms recorded, one should ollow the procedure described in paragraphs 3.1 and 3.2 in particular apply equation (4) in a pixel-by-pixel manner. Fig. 4 (a) presents the image o measured gain non-uniormity distribution. It contains both the low and the high requency component. The ormer appears due to uneven matrix illumination rom inrared emitter, and may be extracted by applying the low-pass iltration - Fig. 4 (b). The high requency component stands or the real gain non-uniormity, and may be calculated by subtracting Fig. 4 (b) rom Fig. 4 (a) the result is visible in Fig. 4 (c). One may notice the issue o vignetting, which is to be addressed in uture works. a) b) c) Fig. 4. a) Measured gain non-uniormity distribution burdened with uneven matrix illumination by inrared emitter, b) the image o matrix illumination intensity distribution by inrared emitter obtained by low-pass iltration o (a), c) measured gain non-uniormity distribution across the microbolometer matrix, with removed low requency component (b) The main source o measurement errors lies in the thermal drit phenomenon, which causes the microbolometer oset shits. Using (5), authors calculated the image o the oset non-uniormity - Fig. 5 (a). There are edge artiacts in the image, possibly because o the sharpening algorithms built into the thermal camera applied or the experiment. In the inal applications, however, raw thermograms are to be used instead o pre-processed ones and thereore such problems shouldn t appear. I gain is assumed to be equal or all microbolometers, the non-uniormity correction becomes one-point the result o such a correction is shown in Fig. 5 (b). The eect o two-point NUC (including both oset and gain nonuniormity) is shown in Fig. 5 (c). I one observes the correction results (Fig. 5 b, c) similarity to the reerence thermogram shown in Fig. 4 (a) is clearly visible. It proves that proposed method works correctly. In case o quantitative analysis o this particular experiment, the mean measurement error introduced by the thermal drit phenomenon was reduced by about 92% with the use o proposed correction method. This value was calculated using ormula (7) as the percentage reduction o mean square error value (MSE) given by (6).
MSE X = M N i= 1 = 1 ( Y REF ( Ti ) X ) M N 2 (6) where: M number o columns in a microbolometer matrix, N number o rows in a microbolometer matrix, Y REF (T i, ) the reerence response o a microbolometer in a position i in a detector (not burdened with the thermal drit) to an IR radiation lux Φ corresponding to the temperature T i,, X stands or: Y(T i, ) the response o a microbolometer in a position i in a detector to an IR radiation lux Φ corresponding to the temperature T i,, burdened with the thermal drit or calculating MSE value introduced by the thermal drit, or φ(t i, ) or calculating MSE value ater the correction with proposed method. EFF = MSE ) MSE MSE ) φ ( Ti ) 100% (7) a) b) c) Fig. 5. a) calculated image o the oset parameter nonuniormity across the microbolometer matrix, b) the result o one-point nonuniormity correction with the proposed method, c) the result o two-point nonuniormity correction with the proposed method 5. Conclusions Authors managed to perorm the two-point nonuniormity correction without interruption o the thermal observation, what was not possible in the case o the traditional shutter-based method. Knowing the transmittance o the ilter (in case o this research τ = 0,48), camera may compensate or it when displaying thermogram, so that the introduction o a ilter may be invisible or the user. Also the IR emitter signal may be eventually iltered out rom displayed thermograms. Proposed method was proven to be useul and eicient or applications such as shutterless measurement thermal cameras. In the inal applications both the inrared ilter and emitter are to be built into the camera, and operate without any user input. It is planned to replace the removable ilter with tunable one, to simpliy the construction and remove mechanical motor otherwise necessary or removing and inserting the ilter. The proposed method is under permanent development. In case o the research presented in this paper, the twopoint correction quality was not any better than one-point approach. However, it is believed that the two-point correction should show its supremacy in case o thermograms with high temperature span. In uture works, authors will maximize the correction eiciency by introducing thermograms averaging and motion estimation techniques.
REFERENCES [1] Orżanowski T., Sosnowski T., Madura H., Bieszczad G., Investigations o response nonuniormity o microbolometer ocal plane array In thermal camera, Pomiary Atomatyka Komputery w Gospodarce i Ochronie Środowiska, no 2, pp. 7-10, 2008 (in Polish). [2] Więcek B., Olbrycht R., New method or on-line microbolometer detector nonuniormity correction, 16th International Conerence on Thermal Engineering and Thermogrammetry (THERMO), Budapest, Hungary, 1-3.07.2009. [3] Więcek B., Olbrycht R., Kastek M., Orżanowski T., Sosnowski T., The method or gain non-uniormity correction o microbolometer detectors (in Polish: Sposób korekci nieednorodności wzmocnienia matryc mikrobolometrycznych ) Patent application no P-387173 dated 02.02.2009 (in Polish). [4] Więcek B., Olbrycht R., The method or elimination o temperature drit inluence on the quality o thermal images generated in microbolometer detectors (in Polish: Sposób eliminowania wpływu drytu temperaturowego matryc mikrobolometrycznych na akość obrazu generowanego w tych matrycach"), patent application no P-389303 dated 19.10.2009 (in Polish).