Comparison of the temperature accuracy between smart phone based and high-end thermal cameras using a temperature gradient phantom

(1)

PROCEEDINGS OF SPIE

SPIEDigitalLibrary.org/conference-proceedings-of-spie

Comparison of the temperature

accuracy between smart phone

based and high-end thermal cameras

using a temperature gradient

phantom

Klaessens, John, van der Veen, Albert, Verdaasdonk,

Rudolf

John H. Klaessens, Albert van der Veen, Rudolf M. Verdaasdonk,

"Comparison of the temperature accuracy between smart phone based and

high-end thermal cameras using a temperature gradient phantom," Proc.

SPIE 10056, Design and Quality for Biomedical Technologies X, 100560D (16

March 2017); doi: 10.1117/12.2252898

Event: SPIE BiOS, 2017, San Francisco, California, United States

(2)

Comparison of the temperature accuracy between smart phone based

and high-end thermal cameras using a temperature gradient phantom

John H. Klaessens

*

, Albert van der Veen, Rudolf M. Verdaasdonk

VU University Medical Center Amsterdam, Department of Physics and Medical Technology

De Boelelaan 117, P.O. Box 7057, 1007 MB Amsterdam, The Netherlands

* j.klaessens@vumc.nl

, www.vumc.nl

ABSTRACT

Recently, low cost smart phone based thermal cameras are being considered to be used in a clinical setting for monitoring physiological temperature responses such as: body temperature change, local inflammations, perfusion changes or (burn) wound healing. These thermal cameras contain uncooled micro-bolometers with an internal calibration check and have a temperature resolution of 0.1 degree. For clinical applications a fast quality measurement before use is required (absolute temperature check) and quality control (stability, repeatability, absolute temperature, absolute temperature differences) should be performed regularly. Therefore, a calibrated temperature phantom has been developed based on thermistor heating on both ends of a black coated metal strip to create a controllable temperature gradient from room temperature 26 ºC up to 100 ºC. The absolute temperatures on the strip are determined with software controlled 5 PT-1000 sensors using lookup tables. In this study 3 FLIR-ONE cameras and one high end camera were checked with this temperature phantom. The results show a relative good agreement between both low-cost and high-end camera’s and the phantom temperature gradient, with temperature differences of 1 degree up to 6 degrees between the camera’s and the phantom. The measurements were repeated as to absolute temperature and temperature stability over the sensor area. Both low-cost and high-end thermal cameras measured relative temperature changes with high accuracy and absolute temperatures with constant deviations. Low-cost smart phone based thermal cameras can be a good alternative to high-end thermal cameras for routine clinical measurements, appropriate to the research question, providing regular calibration checks for quality control.

Keywords: Thermal camera, phantom, quality, smart phone.

1. INTRODUCTION

With thermal cameras the spatial and temporal variations in temperature can be measured and presented as an image. The last years the use of thermography has enormously increased[1,2] and we expect that this will continue in the future. With the development of micro-bolometer array’s without cooling low cost thermal camera’s with a good temperature and spatial resolution became available. This has contributed to an increase in use of thermal cameras in clinical applications. Depending on the research question absolute temperatures or absolute temperature differences or qualitative temperature differences can be measured. For the first two possibilities we need calibrated camera’s if only differentiations in heat spots need to be observed then calibration is not that important.

Thermography is based on the fact that all materials above absolute zero (-273.15 C) emit electromagnetic radiation which is proportional to its temperature[3,4]. The spectral properties of this radiation are described by the radiation law of Planck. The energy per second radiated from a body depends on the surface area and the temperature described by the Boltzmann equation:

= [W m-2] (1) T = T in [K]

σ= Stefan Boltzmann constant = 5.67 10-8 W m-2_K-4

ε = emissivity

A= surface area of body

Proc. of SPIE Vol. 10056 100560D-1 Downloaded From: https://www.spiedigitallibrary.org/conference-proceedings-of-spie on 09 Jan 2020

(3)

The thermal camera collects all the infrared light in the wavelength band of 8-12 µm (Long Wave Infra-Red, LWIR) received from the surface of an object and relates this to a temperature.

The surface properties of the object influences the total infrared energy leaving the object. This can be the emitted radiation based on the objects temperature, the transmitted light trough the object and the reflected light from the surface (Kirchhoff’s Law):

+ + = 1 (2) ɛ = emissivity

= reflectivity τ = transmissivity

The camera sees the total radiation and cannot distinguish between the sources, the operator must give the camera the coefficients of each source to calculate the correct temperature. Metals and tissue are opaque, then only the emissivity and reflectivity at the infrared region influence the temperature measurement. The atmosphere between the object and the camera will also emit and absorb radiation in the IR band depending on the temperature and its composition. The emissivity is a measure of efficiency of a surface to emit radiation relative to a blackbody. A blackbody is theoretically a perfect absorber (all incoming radiation is absorbed) and there is no reflection or transmission, the absorption is equal to the emissivity and is equal to 1.0. All real surfaces have an emissivity lower than 1.0. Emissivity depends on the material, the roughness of surface (polished or rough), wavelength ( MWIR or LWIR range), temperature of material, viewing angle and surface geometry and should be written as: ε (λ,T,ϑ,Φ). The emissivity of a black paint or chalkboard paint (CBP) is very high (ɛblackpaint = 0.9- 0.94, ɛCBP = 0.98) comparable to human skin (ɛhuman skin = 0.98 race and skin

color has no influence). If no emissivity correction is done the observed object temperature is colder than its real temperature.

The resolution of the camera and the spot size of the object in pixels can influence the accuracy of the measurement. Nearby surfaces with deviating temperatures can influence the actual temperature. The spot-size must be at least 10 pixels in diameter for a meaning full temperature measurement and 20 pixels in diameter to give an accurate measurement.

In addition to the correct use of thermography it is necessary to know your thermal camera; how accurate is the absolute temperature measurement, how constant is the temperature registration over time (no temperature drift), how reproducible is the temperature measurement, how accurate is the relative temperature difference (between different areas in the image or at one location at different time stamps) or how correct is the temperature measurement at different distances. This we compare for one high end camera and 3 low-end smartphone based camera’s.

2. METHOD AND EXPERIMENTAL SETUP

2.1 Method

The standard calibration tools for IR thermometers and thermal cameras are blackbody reference sources[5], the uncertainty of these blackbody sources is around 0.2 C, the measurements and equipment are in conformity with the standard ISO/IEC17025:2005. In this study we are not calibrating the camera’s, this study compares the performance of the camera systems using a temperature-gradient tool. It is a practical tool to test the camera’s on performance of absolute temperature absolute temperature differences (gradients), repeatability and stability. These test can be performed on a regular basis to find out if there are possible system errors or calibration errors. If a discrepancy is found between the camera and the gradient tool then the camera can be tested with qualified calibration tool. The idea of this gradient temperature tool is an easy to use testing tool for a quality control of the thermal camera in daily practice.

2.2 Materials

Thermal Camera: FLIR- ONE

The low-end camera in this test is the FLIR-ONE (FLIR systems, Wilsonville, OR, USA), it is a smart phone add-on, in this study the camera is attached to an iPad (Apple, Cupertino, CA, USA). The FLIR ONE contains two cameras, a thermal camera (160x120 pixels) and a visible VGA camera (640x480 pixels) and has a temperature range of -20° to

(4)

120° C with connector for (Vernier Soft Thermal Cam The high-end bolometer an spectral sens software, the Temperature The temperat with an accur aluminum an connected wi Temperature the measured correction va We estimate b 0.5 ºC The standard desired tempe the emissivity The gradient constant temp The gradient gradient syste temperature s over longer ti In this study t of 10 ºC at di Figure 1 Left th the Xenics Gob

a temperatur r long-term m tware & Techn

mera: Xenics d thermal cam d a 384x288 p sitivity LWIR camera has a e gradient ph ture gradient p racy is 0.2

º

C d is back - coa ith a measurin stabilization i d temperature a lue, by which because of res calibration of eratures with a y is known un system is not perature-gradi system produ em has no acti settings. For th imes. temperature g ifferent startin he FLIR ONE t bi 384 thermal c re resolution measurements nology, Beave s Goby 384 mera was a ca pixel array. Th R (8 to 14 μm ethernet inter hantom phantom is ma (figure 2 and ated with blac ng error is 0.2 is done by me and the set va h the heating is

sistance transi f the thermal c a high stabilit nder these con intended to d ients. uces constant t

ive cooling th hese high setti gradients were ng temperature thermal camera camera. of 0.1°C. Th s. The acquisi erton, OR, US alibrated (201 he lens was an m). The data rface to a lapto

ade of two hea 3). The tempe ckboard chalk ºC easuring and c alue. A propor s adjusted. itions between camera is pref ty. The temper ditions the cal o a calibration temperatures o he temperature ings active co e applied in ph es.

a and on the rig

he camera is ition and ana SA). 6) Xenics Go n OPHIR Sup were record op PC, the dat ater-blocks wi erature can be paint with an ontrolling. A rtional−integra n the electroni formed using c ratures are me libration cons n of the IR the or gradients ov e could not be ooling was app hysiologically ght connected to alyzes softwar obi 384 (Leuv pIR 18mm. Th ded and analy ta were acquir ith thermistor e set from 26 º n emissivity of feedback syst al−derivative ( ic parts that th calibrated rad easured with a tants for the in ermal camera

ver the length kept constant plied with a fa possible temp an external re applied is ven, Belgium) he sensor has yzed with Xe red with 25 Hz temperature f ºC to 100 ºC. T f 0.98. On this tem minimizes (PID) controll he accuracy of diation sources a high precisio nfrared sensor but it produce h of the black m t over longer t an this gave sta

perature chang battery using Vernier Ther ), with an un 16 bit dynam eneth64 (Xen z. feedback cont The metal stri s strip five Pt1 s the differenc ler algorithm f the total syst s, these can pr on calibrated t rs can be dete es constant tem metal strip. Be time periods a able temperat ges, we applie g a mini USB rmal Analysis cooled micro mic range and a

nics, Belgium) rolled system ip is made of 1000 are ce between calculates a tem will be roduce the thermometer, rmined. mperatures or ecause the at high ture gradient ed a gradient B s -a )

(5)

1 1 1 1 1

\

N

1 1 1 1 2.3 Measure A compa • • • • • • The mea heating s Figure 4 Expe metal plate wi Heating bl 40 C 30 C Figure 2 S th ements and se arison is made Uniformity o Repeated mea measurement Absolute tem Absolute tem Temperature Temperature surements set sources and pl erimental setup ith constant tem

10.5 cm lock Schematic draw hermal image o etup e between the f the sensor asurements, th ts. mperature mperature diffe relation to dis gradient, twee tup are shown

aced in a shie

with a flat blac mperature (emis 1.5 cm 1.5 cm Pt 1000 wing of gradien of a gradient of high-end and he camera is s erence stance (Object e gradients ov in figure 4 an elded environm

ck coated ssivity 0.98). He t phantom and f 10 degrees Cel low-end IR th switched off fo t distance) ver 10ºC from nd 5, the whol ment to preven IR camera IR Thermal cam Figure 5 Expe phantom. 1.5 cm eating block Fig lsius. hermal camera or 1 minute be m 30 ºC -40 ºC le system is pl nt reflections f D = o mera erimental setup gure 3 Tempera on the b painted a on the follow etween the dif

and from 40 º laced in a room

from other sou

object distance

with the tempe ature control-un ottom an image metal strip wing topics: fferent ºC -50 ºC m with no oth urces. Temperature gradient phan erature gradien nit (top), e of the black her tom

t

(6)

3.1 Uniform The uniformi at constant ro The cameras The field of v cameras is sh Analyses are The average o camera 23.8 ± 0.4 ºC. A mea camera is this 3.2 Repeata The repeated the measurem Figure 6 On t (B) is shown b the low end cam

mity over sens

ity of the senso oom temperatu

are placed at view (FOV) of hown in figure

performed in over the whol ±0.3 ºC. In on asure for the s s 0.2 ºC and f

ability

temperature m ments the equip

top is an image below and the a mera on the rig

sor

or is measured ure (setup figu a distance D = f the cameras e 6. 2 ROI’s: (A) e sensor (RO ne image a var stability over t for the low-en

measurements pment is switc

of the sensor w average over the ght site of the fig

3.

d imaging a fl ure 4).

= 30 cm and th see only the b

a square area OI A) in one im riation over th time (minutes nd camera ~ 0. s are preforme ched of for at with a temperatu e whole screen gure.

. RESULT

lat black coate he black meta black metal pl

a over the who mage is for the he diagonal B i s) is the standa .6 ºC. ed using the s least 1 minut ure range of 2 is calculated in

TS

ed metal plate al plate has a t late, an examp

ole sensor and e high-end cam

is for the high ard deviation i etup of figure e. ºC. The tempera n area A. The hi (emissivity in emperature of ple of the temp

(B) over the d mera 23.2 ±0.3 h-end camera in the average e 4, the distanc ature variation igh end camera

n the camera i f 22.2 ± 0.5 ºC perature imag diagonal of th 3 ºC and for th 1.1 ºC and for e in area A; fo ce is D=30 cm

over the diago a is on the left a is set to 1.0), C. e of both he sensor. he low-end r the low-end r the high-end m. Between al onal and d ll

(7)

We observe t cameras. If w but measure a start-up effec 0.35 and 0.35 3.3 Tempera The experime phantom is se The temperat at position 1, The results ar (thermal cam and apparent Celsius lower a higher appa gradients. The set absol measures a te cameras have

Figure 7 The d real temperatu

that the first m we leave the fir at a constant t ct) is for the hi 5 ºC. The temp ature gradien ental setup is a et in the physi tures are meas 3 and 5 in fig re presented in meras) are fitted temperatures. r for the 40-5 arent temperat ute temperatu emperature de e a lager devia deviation of the ure. All the four

measurement d rst measureme emperature le igh-end Xenic poral variation nt measurem according to f iological temp sured with 5 P gure 8. n figure 8, the d with a 2e or . The apparen 0 ºC gradient. ture. The appa ure differences ference within ation of 14, 12 e measured (app tested cameras deviate much f ents out we fin evel. The stand cs camera 0.7 n in temperatu ents figure 5, the ob perature range Pt1000 sensors e real (phantom rder polynomi t temperature . The three FL arent and real

s over the met n the accuracy 2 and 5 %.

parent) tempera s are presented

for the later on nd still that al dard deviation

ºC and for the ure within eac

bject distance of 30-40 ºC a s of the phanto m temperature ial function. T is 3 degrees C LIR-ONE cam temperature c tal strip of the y of the phanto

ature from the d separately.

nes this can be l cameras hav n of the variati e low-end FLI h measuremen e D= 30 cm. T and at a higher om and at 3 R e measuremen The Xenics cam

Celsius lower mera’s differ fr curves are not phantom is 1 om, a deviatio e induced to s ve a deviation ion in apparen IR-ONE came nt is small: 0.0 The temperatur r temperature ROI on the strip

nts) and the ap mera shows pa

for the 30-40 rom each othe parallel over 0 ± 0.5 ºC (fig on smaller tha

start-up effects from the real nt temperature eras respective 05 – 0.1 ºC.

re over the gra range 40-50 p with the the pparent temper arallel curves ºC gradient a er, 2 give lowe

both the temp gure 9). The X an 5%. The FL s of the temperature e (without ely ± 0.13, adient ºC. ermal cameras ratures

for the real and 1 degree er and 1 gives perature Xenics camera LIR ONE s a

(8)

1

Figure 9 The a and for the rep Figure 8 The a temperatures m order polynom 3 2 absolute temper peated measurem apparent tempe measured with t mial function. 4 rature differenc ments. The erro eratures measur the Pt1000 on th

5

ces between poi or bars give the red with the fou he gradient pha

1

int 1 and 5 are p e standard devia ur thermal came antom (dotted l 3 2 presented for th ation within eac eras (solid lines ines). All curve

4

he four camera ch measuremen s)and the real es are fitted with

5

as nt.

h 2e

(9)

3.4 Tempera We use the ex the distance D camera’s both The Xenics c the phantom. The FLIR-ON measurement is in accordan In clinical the the measurem measurement this can be do This study sh end camera a The absolute pyrometer. Th the plate after temperature i confirmed wh change. The FLIR-ON camera may n compared the differences be Repeated abs temperature c Figure 10 The function of the 26 ± 0.5 ºC. ature measur xperimental se D is varied fro h cameras are amera gives o NE 1 and 2 ca t range of 10 c nce with the p

ermal imaging ments are perfo ts both camera one within one hows that the t

nd 0.4 C for th temperatures he constancy r a long time ( is constant thr hen the plate i NE camera can not be switche e FLIR-ONE i etween health solute measure comparisons o e apparent temp object distance rement in rel etup of figure om 10 to 200 c aligned at eac

over the object amera give wit cm – 150 cm, phantom. The F g mostly absol ormed in one a are stable ov e image, betw temperature se he low-end ca of the gradien and homogen (> 24 hours) w oughout the m is slightly mov n measure cor ed off between is as accurate hy and disease ements (chapt over short time

perature from th e. The phantom ation to the o 5, the temper cm. The objec ch object dista t distance rang thin the accura the absolute t FLIR-ONE -3

4.

lute temperatu ROI on the se ver a time peri ween different ensitivity at di amera. nt phantom sy neity of the bla was at room te measurement, ved (left-right rrect temperat n measuremen as the high-en d tissue are co er 3.3) show o e periods, the he four thermal temperature is object distanc rature of the m ct is always pu ance.

ge of 20 – 150 acy range of t temperature is 3 gives a large

DISCUSSI

ure difference ensor. This req iod ( 0.2C for

areas of the se ifferent region ystem was veri ack coated me emperature. Th

and there wer t), we observe ture difference nts over time ( nd camera. Th ompared[6]. offsets in temp cameras may l cameras as s constant ant ce metal strip is se ut in focus of t 0 cm the corre the phantom (± s shifted for ca e spreading in

ION

s are measure quires stable c the high end a ensor. Now th ns can differ m

ified with exte etal plate (figu

his is most lik re no heat sou ed that the tem es as they are (chapter 3.2). his is of impor perature. This y not be switch et to a constan the Xenics cam

ect temperatur ±0.5 C) const amera 1. For c temperature. d. This can be cameras, as we and 0.6C for t he uniformity more than 1 de ernal thermoc ure 4) can be q kely the case b rces close to t mperature imag small (< 5 ºC If temperatur rtance in studi s is important hed off. nt level of 26 mera, for the F

re within the a ant results in t camera 2 the a e done over a e have seen in the low-end ca of the sensor egree Celsius couples and a questioned. W because the ro the black plate ges of figure 6 ) in chapter 3. es within one es were e.g. te when studyin ± 0.5 ºC and FLIR-ONE accuracy of the absolute value time period, n the amera). Or is important. for the

high-calibrated We assume that om e. This is also 6 did not .3, the image are emperature g absolute e t

(10)

To know your own camera the measurements in chapter 3 should be performed. The three FLIR-ONE cameras give different results and camera number-3 gives large deviations, this camera was one of the first of this generation and can possibly differ in hard and software. All the four cameras give small variations in temperature over time, the test measurements were done for only several minutes. In future studies the stability should be checked over longer time periods. The analyzing software for the high-end Xenics camera is much more sophisticated than the Apps of the FLIR-ONE camera. This limitation may hinder the deployment of these low-end cameras. The temperature – object distance variation show that for clinical relevant distance ranges (20-150 cm) the Xenics and 2 FLIR-ONE cameras give good results.

5. CONCLUSION

A summary of the findings in this study are shown in the table below.

High End Xenics Gobi 384 Low –End FLIR-ONE

Setup system - Connection laptop + tablet of iPhone Analyzing software + more sophisticated - Limited tools

Spatial resolution + 384x288 - 160x120 Spatial temperature variation

over sensor - Vary up to 1.0 ºC + Vary up to 0.4 ºC Temporal variation within

one measurement + whole sensor 0.2 ºC _{+ ROI temporal stdev 0.05-0.1 ºC} - Whole sensor ~ 0,6 ºC _{+ ROI temporal stdev ~ 0.2 ºC} Absolute temperature + constant deviation

+ up to 1 ºC deviation

- Constant deviation - Up to deviation 3 ºC Repeated measurements + vary < 0.7 ºC + vary < 0.35 ºC Gradient step 10 ºC + constant deviation from real

temperature

- Variation in deviation from real temperature

Δ Absolute temperature + small < 0.5 ºC (< 5%) (within variation of phantom)

- < 14, 12 and 5% deviation of real T ( for real step T = 10 ºC) Object distance + 20 - 150 cm < 0.5 ºC + 10-150 cm < 0.5 ºC

(one camera not correct)

Both low-cost and high-end thermal cameras measured relative temperature changes with high accuracy and absolute temperatures with a deviations from the real temperature. Low-cost smart phone based thermal cameras can be a good alternative to high-end thermal cameras for routine clinical measurements, appropriate to the clinical research question, providing the regular calibration checks for quality control and strict protocols should avoid the pitfalls shown in this study.

(11)

6. REFERENCES

[1] Hildebrandt,C., Raschner,C., and Ammer,K., (2010). An overview of recent application of medical infrared thermography in sports medicine in Austria. Sensors. (Basel) 10, 4700-4715 (2010).

[2] Ring F., Jung A., and Żuber J., Infrared Imaging, A casebook in clinical medicine, IOP Publishing Ltd (2015) [3] Nicholas A. Diakides, Joseph D. Bronzino, Medical Infrared Imaging, CRC Press (2008)

[4] Kaplan H., Practical Applications of Infrared Thermal Sensing and Imaging equipment, SPIE Press (2007) [5] Machin G.,Simpson R., Broussely M., Calibration and validation of thermal imagers, 9E International

Conference on Quantitative Infrared Thermography (2008)

[6] Jaspers,M.E., Maltha,I., Klaessens,J.H., de Vet,H.C., Verdaasdonk,R.M., and van Zuijlen,P.P., Insights into the use of thermography to assess burn wound healing potential: a reliable and valid technique when compared to laser Doppler imaging. J. Biomed. Opt. 21, 96006 (2016).