2021 International Conference on Nascent Technologies in Engineering (ICNTE 2021)

Enhanced Pyrometric device with Long Range for

mass screening based on MLX90614
Dishant Shah Pooja Jha Vedant Awasthi
2021 4th Biennial International Conference on Nascent Technologies in Engineering (ICNTE) | 978-1-7281-9061-7/21/$31.00 ©2021 IEEE | DOI: 10.1109/ICNTE51185.2021.9487689

B.E., Electronics and Telecommunication Data Engineer B.E., Electronics and Telecommunication
Dwarkadas J. Sanghvi College of Engineering Quantiphi Incorporation Dwarkadas J. Sanghvi College of Engineering
Mumbai, India Mumbai, India Mumbai, India
[email protected] [email protected] [email protected]

Parth Mau Bhavin Kothari Isha Maru

Senior Data Analyst Graduate Student in Management Information Systems Analyst Deloitte
Ganit Incorporation University of Maryland College Park Mumbai, India
Mumbai, India Maryland, USA [email protected]
[email protected] [email protected]

Abstract—This paper describes a novel data analysis algorithm like - airports, malls, public places, and industries. However,
for increasing the range and accuracy of non-invasive IR based thermal cameras were extremely expensive, which was a
temperature sensors that have been widely used to check for letdown for every shop and company owner that wanted to
elevated body temperatures (EBT) during the Corona pandemic.
Our algorithm considers factors like humidity, distance, ambient resume work and production but could not afford this kind of
temperature, and emissivity. A device was made that could investment, especially due to the widespread economic
deploy the algorithm and perform tasks like EBT detection, mask recession that followed the pandemic. This paved the way for
detection, face detection and face recognition and subsequently cheap thermal guns that claimed to sense body temperatures to
store the inferences of all these models on a cloud or local sever an accuracy of 0.2 degC which became extremely popular
accessible via http requests.
Index Terms—Elevated Body Temperature, EBT, IR, Infrared, with such customers and industrialists. However, these devices
Artificial Intelligence, AI, face recognition, thermal screening, had a lot of problems - firstly, they were extremely short
COVID-19, pandemic, data analysis, MLX90614, NodeMCU, ranged which resulted in a lot of manual measurement errors
ESP32 cam as it is quite natural for the user to be cautious about the
distance between the gun and the subject’s forehead to avoid
I. INTRODUCTION any possibility of contracting the virus. Secondly, the accuracy
In December 2019, the SARS-CoV-2 outbreak began in of spot IR devices has always been a concern due to the ample
Wuhan, China. As time passed, it began to infect people in of factors affecting it, including but not limited to relative
other countries and by June 2020 more than 200 countries humidity, Ambient temperature and emissivity. This combined
had positive cases amounting to over 10 million reported with the introduction of indistinguishable cheap quality
infections and over 500 thousand reported deaths due to the sensors to increase profit margins per device made these
virus all over the world. As pressure began to mount on each devices notorious and speculative among many users and
infected country after its first few infections were identified, researchers. The main purpose behind this research is to
widespread Coronavirus testing methods were employed to address all these concerns, identify the main factors that affect
make sure infected people were identified and quarantined the EBT measurement and their relationship with it; and
before they infected someone else unknowingly. The most develop an algorithm that incorporates all these factors
prominent methods of detecting the virus included chemical (including the distance between the subject and the spot
methods like RT-PCR, which were time consuming and labor detector) to increase the range to at least 1 foot. Finally, we
intensive. Other faster and more automated methods like aim to deploy this algorithm by the means of our own custom-
detection using AI and Deep Learning models based on X- made cost- effective device that can run other simple models
Rays, CT-Scans and thermal imaging modalities were being like face recognition, mask detection and track attendance of
explored, but they were still at nascent stages and not ready for workers. All this data can be locally saved, and the inferences
mass adoption and deployment. Soon, a very strong need for based on these algorithms will be logged on a remote or local
screening people who showed mild symptoms was needed so server, accessible via http requests or other methods.
that the burden on test centers and medical facilities is
reduced. Thermal imaging and EBT detection were the most II. LITERATURE SURVEY
widely employed screening method that was used by a variety There are a lot of ongoing efforts to develop thermal
of countries at places with large public densities and diversity detectors or pyrometers currently. Unfortunately, not many

978-1-7281-9061-7/21/$31.00 ©2021 IEEE

Authorized licensed use limited to: PES University Bengaluru. Downloaded on January 18,2023 at 12:01:54 UTC from IEEE Xplore. Restrictions apply.
 2021 International Conference on Nascent Technologies in Engineering (ICNTE 2021)

papers exist which elaborate on making the measurements with timestamps and other necessary data like the person’s
more accurate by taking actual measurements and curve fitting temperature, his attendance, etc.
them to estimate the error at unknown distances or ambient
temperatures. One of the papers that caught our interest was [1]
that elaborated on using a cheap STM32F107 microcontroller
to run the MLX90614 pyrometric sensor. It had a data storage
system and an alarm/LED for issuing warnings. However, the
paper did not include any extra features like surveillance, face
detection, etc. in the device. The paper also fails to show the
performance of the device on humans, which is our main
concern. The second paper that we referred was [2] which uses
a low-resolution thermal camera from the same company
Melexis, MLX90640. It was useful to understand the process
of sending visual data to a cloud server via MQTT protocol
as elaborated in the paper. However, we are using MLX90614
pyrometer and a 2MP visible camera for surveillance, making
our device cost much lower than theirs. Finally, [3] showed
the use of esp8266 based devices to send data on a server.
Most of these papers mainly focus on developing a simple
pyrometer that can send data to a cloud server. We aim to Fig. 1. Assembled PCB showing all components
develop a surveillance system that can perform pyrometry as
well as basic face detection. All the inferences will then be sent 3. Camera sensor: The ESP32-CAM device contains an
on the cloud server, making our device much more versatile, inbuilt camera port which can be used to connect the OV2640
remotely operable and feature rich. Our main aim is to make it camera sensor. A cost-effective highly customizable camera, it
as affordable as possible for mass usage and adoption. is ideal for basic image processing and surveillance
III. C OMPONENTS USED AND METHODOLOGY applications. It supports up to 2MPx of resolution and still
gives decent frame rates.
The device comprises of the following blocks and
4. Buzzer: A small 5V buzzer has been used to provide audio
components as shown assembled in Fig. 1 -
cues and output in the device. It acts as an alarm or guidance
1. Pyrometric sensor: The MLX90614 has been a prevalent
system as per requirement. When a person has elevated body
choice as the main sensor. It has a lot of variations, one of
temperature, or is not recognized by the database (intruder
them being specifically calibrated to achieve better accuracy
alert) then the buzzer sounds an alarm. It also helps the person
in medical applications (MLX90614DAA). However, these
to align his/her forehead to the pyrometer by guiding the
variants are not easily found and are priced higher. Also,
subject to come closer or further from the device for safe and
according to the datasheet, for normally occurring ambient and
accurate measurement.
object temperatures the accuracy of all the variants is quite
5. Display: A 0.96” to 1.3” display can be used to show the
similar, around 0.3-0.5 degC according to the datasheet as
temperature values and other data to the subject after his
shown in the figure 1. Hence, we decided to use the most
measurement is complete. It is an I2C display and is extremely
commonly available market variant so that our device can be
easy to interface with the remaining hardware.
reproduced easily. The MLX90614 is an I2C pyrometric
sensor from the company Melexis, a leading manufacturer of 6. Ultrasonic Sensor: The SR-04 Ultrasonic sensor has been
low-resolution IR sensors. It gives very accurate temperature used to measure the distance between the subject and the
readings at 1-3 cms separation from the object being measured. device. It is used as a parameter to compensate the error
However, under practical circumstances, as the distance is introduced by the influence of ambient temperature in the
increased beyond that the measured values for the same human object temperature measurement by the pyrometer as the
subject change as the influence of the ambient temperature of distance between the device and the subject increases.
the sensor’s immediate surroundings increases on the object 7. Power circuit: A simple TP4056 based circuit has been
readings measured by it. We aim to improve this measurement used to provide sufficient power to the entire device
over a much larger range of distance around 30 cms or 1 foot. electronics. It can be powered by a micro-USB cable.
This is elaborated in the further sections. Construction: The entire device casing is built using 3D
2. Wifi module: The ESP8266 based Node MCU module has printed parts and locally available electronic parts. It is a
been used to provide Wifi facilities, temperature compensation simple to use snap fit 2-piece design made of a front and
and commands to camera module. It acts as the master device. a back part. The device is mounted on a wall using a few
The image from ESP32 Cam will be input to various machine additional 3D printed parts.
learning models to detect faces, masks and humans. The Use: The block diagram of the entire device has been
inferences will be stored on a cloud server using Wifi elaborated in the Fig. 2
978-1-7281-9061-7/21/$31.00 ©2021 IEEE

Authorized licensed use limited to: PES University Bengaluru. Downloaded on January 18,2023 at 12:01:54 UTC from IEEE Xplore. Restrictions apply.
 2021 International Conference on Nascent Technologies in Engineering (ICNTE 2021)

The ambient temperature affects our body temperature and

hence, our body is made to regulate environmental temperature
changes to keep our body temperature fixed. Unfortunately,
this might cause fluctuations in the IR rays emitted by our
body, so it is recommended to let the sensor and the body
stabilize and adjust to the environment for 15-30 mins and
then start measurements. The ambient temperature also affects
the above IR v/s distance graph, as shown in Fig. 3
Hence, we make an entire dataset encapsulating all these
changes in csv format and feed it to MATLAB’s curve fitting
function and perform error compensation as elucidated in the
Fig. 2. device Block Diagram following section.
V. ANALYSIS AND RESULT:
IV. P RINCIPLE AND CONCEPT:
Thermopile based temperature measurement works on a
simple principle illustrated by the Stefan Boltzmann’s law
- Hot bodies emit radiation, in the form of IR waves. The
quantity of radiation emitted, and its relation with temperature
of the emitting body are given by the Stefan Boltzmann’s law:
P = AsσT 4 (1)
where,
P: Power emitted by the object emitting the radiation (in
Watts)
A: Surface Area of the object emitting the radiation (in
metre2) Fig. 3. Effect of the distance between the pyrometric sensor and the subject to
be screened for various ambient temperatures ranging from 73 degF to 91
s: Emissivity of the object emitting the radiation (constant degF
unique to the object properties, no units)
σ: 5.670373× 10-8 (W.m.) To observe the effects of ambient temperature and distance
T: Temperature of the object emitting the radiation (in on the pyrometer readings, we first collected data from the
Kelvin) pyrometric sensor, using it to measure a subject’s forehead
temperature, taking a mercury thermometer reading of the
Using this relation, it is possible to back-calculate the surface forehead point exactly in front of the sensor as the ground
temperature of a non-shiny object, like the human skin or a truth and varying the distance between the subject and the
vessel of hot water using the inputs emissivity, IR rays pyrometric sensor at different ambient temperatures. We
received from a unit area of the scene and their intensity. recorded all our data in an excel workbook. We then plotted
However, in practical situations, due to attenuation loses the the collected data on Microsoft PowerPoint for a very basic
distance between the object emanating IR rays and the IR visualization as shown in fig. It clearly showed a drop in the
sensor capturing them affects the temperature back calculated sensor reading with increase in distance. We also wanted to
by the sensor. Other parameters that affect the thermal readings plot all the ambient temperatures together, however the
include Ambient temperature, Relative Humidity and lighting distances recorded by the ultrasonic sensor were accurate to
conditions (if the light source emits a lot of thermal radiation, millimeters and there were multiple readings taken within just
like a bulb). 1 cm. To make the graphs scaled and easier to compare, we
To compensate the effects of these factors, we created a implemented a MATLAB script to separate out the data for
compensation function in MATLAB. We collected the temper- each ambient temperature and average out all readings taken
ature readings for different ambient temperatures and varied within a centimeter of distance from each other. Then we
the distance between the sensor and the subject. Using this plotted all the averaged-out data together providing us with a
data and MATLAB functions, we were able to curve fit this very intuitive graph as show in figure 3. Clearly, as ambient
data and applied an error compensation method. The IR temperature decreases, the graph shifts downwards. Using this
temperature readings of the sensor are most accurate when the data, we curve fitted a surface plot in MATLAB using its
distance between the subject and the sensor is less than 1cm. curve fitting tool. We used a bivariable polynomial p42 with
Hence, we interpolate the sensor reading at say 20 cm distance distance and ambient temperatures as the two input variables
to that for 1 cm distance as shown in fig by compensating the and the average value of the sensor at that distance and
error between the values on the curve at those two distance ambient conditions as the output to curve fit the data as shown
values. in Fig. 4.
978-1-7281-9061-7/21/$31.00 ©2021 IEEE

Authorized licensed use limited to: PES University Bengaluru. Downloaded on January 18,2023 at 12:01:54 UTC from IEEE Xplore. Restrictions apply.
 2021 International Conference on Nascent Technologies in Engineering (ICNTE 2021)

Fig. 5. The graph showing agreement between ground truth (Thermometric

Fig. 4. cftool curve fitting for our data in MATLAB
readings) and compensated sensor measurements (Pyrometric readings)

Using this tool, we got the following p42 polynomial: VI. C ONCLUSION :
2
f (x, y) = p00 + p10x + p01y + p20x + p11xy + ... (2) In this manner, we have increased the sensor’s range from
3cm to 30cm with a good accuracy of +/-0.5 degC when tested
where,
on a batch of 50 subjects of different localities. We aim to
x: perpendicular distance between the pyrometric sensor and
productionize this device using a robust metal casing as shown
the forehead plane (in centimeters)
in the Fig. 6 so that it can be deployed on a large scale to test
y: Ambient temperature (in degC)
people at their work places and recreational/travelling hotspots
f(x,y): function that correlates the variation in temperature
like malls, airports, train stations, etc. This device is also
readings of the pyrometric sensor MLX90614 with change in
capable of surveillance and mails the alerts and temperature
distance and ambient temperature.
values on a discord channel or email address of the concerned
The constants that were calculated using MATLAB are
authorities. Hence, it keeps a record of all the collected data
collected in the Table 1 below.
TABLE I
unlike thermal guns. We also aim to measure more vitals like
VALUES OF CONSTANTS FOR P42 EQN. 2 SPO2 levels, Respiratory Rate (RR) and Heart rate (HR) using
visible and thermal camera feed as explained in [4],[5],[6] and
p00 57.14 p10 -17.67 [7].
p01 0.8357 p20 0.7385
p11 0.3206 p02 -0.004624
p30 -0.002144 p21 -0.01438
p12 -0.001409 p40 -5.051e-05
p31 5.273e-05 p22 6.46e-05

Using this equation and its constants, we can interpolate the

sensor value at 1 cm distance at that ambient temperature at
which the pyrometer is most accurate, i.e., at distance less than
1cm. Say we know the function output at distance x and 0

ambient temperature y (given directly by the pyrometric

0

sensor)
0 , 0 00 10 0 01 0 20 ⋯ (3)
0, 0 00 01 0 02 (4) Fig. 6. Envisaged production ready version of the proposed device with sample
text
0, 0 0, 0 (5)

Now all we have to do is compensate this error by adding it REFERENCES

to the sensor’s object temperature reading. Say, we have the [1] G. Jin, X. Zhang, W. Fan, Y. Liu, and P. He, “Design of non-contact infra-
erroneous object temp value ’ at for ambient temp. y0. The red thermometer based on the sensor of mlx90614,” The Open Automation
compensated object temperature will be: and Control Systems Journal, vol. 7, pp. 8–20, 02 2015.
[2] G. Spasov, V. Tsvetkov, and G. Petyrova, “Using ir array mlx90640 to
0 ’ (6) build an iot solution for all and security smart systems,” pp. 1–4, 09 2019.
[3] G. Marques and R. Pitarma, “Non-contact infrared temperature ac-
t0 is the compensated object temperature. We took a batch of quisition system based on internet of things for laboratory activities
monitoring,” Procedia Computer Science, vol. 155, pp. 487–494, 09 2019.
50 patients to test our formula after implementing it on the [4] Aleš Procházka, Hana Charvátová, Oldřich Vyšata, Jakub Kopal and
NodeMCU using Arduino IDE. We recorded the thermo- Jonathon Chambers, “Breathing Analysis Using Thermal and Depth
metric measurements of each patient before t he measurement Imaging Camera Video Records”
[5] Dingliang Wang, Xuezhi Yang, Xuenan Liu, Jin Jing, and Shuai Fang,
and used our device to measure the compensated pyrometric “Detail-preserving pulse wave extraction from facial videos using
temperature values. We observed a good accuracy of +/- 0.5 consumer-level camera”
degC as shown by the graph in Fig. 5 [6] Ufuk Bal, “Non-contact estimation of heart rate and oxygen saturation
using ambient light”
[7] http://people.csail.mit.edu/mrub/vidmag/

Authorized licensed use limited to: PES University Bengaluru. Downloaded on January 18,2023 at 12:01:54 UTC from IEEE Xplore. Restrictions apply.