Detection and Tracking of the Human Hot Spot

Carlos M. Travieso

, Malay Kishore Dutta

, Jordi Solé-Casals

and Jesús B. Alonso

Signal and Communications Department, The Institute for Technological Development and Innovation on

Communications, University of Las Palmas de Gran Canaria, Campus Universitario de Tafira, sn,

Ed. de Telecomunicación, Pabellón B, Despacho 111, E35017, Las Palmas de Gran Canaria, Spain

Department of Electronics and Communication Engineering, Amity School of Engineering & Technology,

Amity University, Noida, India

Computer Science Department, Univesitat of Vic,Victoria, Spain

Keywords: Soft-biometrics, Human Detection by Hot Spot, Thermal Image, Pattern Recognition.

Abstract: This work presents an algorithm that receives as input a stream of thermal imaging detects heat sources

present in them and classify them according to their mobility. After performing the experiment and given

the results, it is concluded that the algorithm performed is able to discern and classify the different types of

human bodies as long as you can provide a set of detection parameters adjusted to the situation, indoor or

outdoor; and with one or more persons.

1 INTRODUCTION

Capturing images in the infrared spectrum and its

subsequent processing is as diverse as the detection

of forest fires (Liew et al., 2005) areas (Vodacek et

al., 2005) applications of hot spots in electrical

networks (Ishino, 2002) (possible indication of a

short circuit, as it is shown in figure 1) of buried

landmines (Carter et al., 1998; Azak et al., 2003) and

focusing a bit on people, face recognition (Heo et

al., 2005; Socolinsky & Selinger, 2004; Jiang et al.,

2004) of facial expressions and gestures (Jiang et al.,

2005) of positions (Iwasawa et al., 1998; Imai et al.,

1993) or even biometric features such as the

arrangement of veins in the hand (Lin and Fan,

2004).

Figure 1: Hot spots in a grid.

Infrared detection supplements (and in some cases

could replace) conventional video surveillance

because, while relatively easy to hide in the eyes of

others, not so much thermal radiation hide all bodies

emit. As mentioned , the present design aims to

design a system to detect heat sources in a scene and

which is able to track the same, by means of a CCD

camera operating in the IR band.

Although the heat source can be anything that is

at a certain temperature, the application will be

developed oriented to the detection and tracking of

people, making it a system of monitoring and / or

control regions.

A surveillance system of this type could be used

to monitor the check-in counters at an airport, access

to large areas like stadiums or shopping malls,

parking for an office or a building block, the waiting

section hospital emergency ... in general, all those

places with large crowds.

In this project the development of an algorithm

that, starting in the infrared spectrum captured

scenes then rated heat sources present pursued. This

classification is based on whether the foci remain in

the same position and in the same way all the time or

not.

2 USE OF THE THERMAL

SENSOR

To capture thermal images and the construction of

the database has been used a video camera company

Guangzhou SAT Infrared Technology Co. (SATIR),

325

Travieso C., Kishore Dutta M., Solé-Casals J. and B. Alonso J..

Detection and Tracking of the Human Hot Spot.

DOI: 10.5220/0004941703250330

In Proceedings of the International Conference on Bio-inspired Systems and Signal Processing (MPBS-2014), pages 325-330

ISBN: 978-989-758-011-6

 2014 SCITEPRESS (Science and Technology Publications, Lda.)

SAT-280 model (see figure 2). This camera comes

standard with features like interesting presentation

profile or isothermal heat. It is capable of making

measurements at specific points of the stage,

designated by the user, or in areas of rectangular

profile. However, regarding the development of the

algorithm, only interested in the image.

Figure 2: Used IR camera and capture a person.

Regarding the capture settings , it was decided to

establish an ambient temperature of 25°C and

relative humidity was set at 0% because the idea was

to study and common everyday situations. Since it is

expected to find hot spots at different distances,

about the parameter (default is 1 meter) was

avoided. The emissivity of an object or surface is the

ratio of thermal radiation that is capable of

absorbing or emitting from that of a black body

equal and varies between 0 and 1, the latter being the

value for a perfect thermal emitter (Cromer, 1985).

The human body has an emissivity of about 0.97 and

does not change only the color of the skin. Under the

conditions of employment provided, not some

variation was observed in the imaging parameter to

modify the emissivity so it was left to the default

value, unity.

Presentation settings were used regularly the

LEVEL and SPAN as the needs of the algorithm did

not include regions of interest previously

established. Neither wanted to do prior segmentation

temperatures.

Table 1: Daytime measures, focus at 3 meters.

Distance

(m)

Focus

LEVEL

(ºC)

SPAN

(ºC)

Comment

3 * 27 13 Acceptable

5 NO 27 13 Acceptable

8 NO

Confusion

with ground

Acceptable

10 NO 30 13 Acceptable

15 NO

Confusion

with ground

Acceptable

Before capturing images, tests were performed in

an outdoor environment during the morning.

Table 2: Daytime measures, focus at 15 meters.

Distance

(m)

Focus

LEVEL

(ºC)

SPAN

(ºC)

Comment

3 NO 30 13 Regular

5 YES 30 13 Acceptable

8 YES 30 13 Acceptable

10 YES 30 13 Acceptable

15 *

Acceptable

Confusion

with ground

And then tests were conducted in an indoor

environment.

Table 3: Measures night, focus at 3 meters.

Distance

(m)

Focus

LEVEL

(ºC)

SPAN

(ºC)

Comment

3 *

Confusion

with ground

Acceptable

5 NO 30 13 Acceptable

8 NO 30 13 Acceptable

10 NO 30 13 Acceptable

15 NO 30 16 Acceptable

Table 4: Measures night, focus at 15 meters.

Distance

(m)

Focus

LEVEL

(ºC)

SPAN

(ºC)

Comment

3 NO 30 13 Acceptable

5 NO 30 13 Acceptable

8 YES

Acceptable

Confusion

with ground

10 YES 30 13 Acceptable

15 *

Acceptable

Confusion

with ground

Finally it was decided to perform detection based on

the fact that the algorithm would be used in

potentially changing scenes, that is, with hot spots

appearing and disappearing of them, in the most

general case. So to develop the algorithm is first

recorded sequences of frames and then some videos.

This has been referred to a frame sequence number

of images relating to a scene. They are distinguished

from the videos because the rate of frames per

second in the sequence of frames is less than one.

Based screening uptake levels of certain temperature

and changing information between keyframes is

showed when constructing an algorithm with

satisfactory results. The camera has its own memory,

but it could just shoot snapshots , so that it became

necessary to connect the video output of the camera

to a video capture card. The card used is a PCMCIA

Card Imaging, Impex Inc., model VCE- B5A01 and

BIOSIGNALS2014-InternationalConferenceonBio-inspiredSystemsandSignalProcessing

326

is connected to a laptop Aspire 1600 Series, MS2135

model with Windows XP. The card software

includes a simple program that can display the

contents of the video input card. The captured video

can be saved as sequences of frames or video file. In

a line of work sequences of frames in which scenes

were captured at a rate of frames per second less

than unity, ie that passed between frames over a

second were used, about a second and a half. These

sequences, while useful to some extent, were

eventually discarded and finally recorded a few

videos in AVI format at a rate of 14 frames per

second. Mainly due to disk space limitations, short

sequences of less than one minute duration were

recorded.

Eight thermal videos were recorded with the

described capture device. Indoor and outdoor

scenarios were used; and the focus goes in one

motion. The main characteristics of each video are

summarized in the following table 5.

The first column gives the name of the video in

code, in the second column duration (time) is given

in seconds and the third a brief description.

The videos were recorded during daylight hours.

The camera was focused on objects eight to ten

meters. The LEVEL and SPAN values were set to

(L: 30, S: 13). The format is AVI videos, all have a

rate of 14 frames per second and a size of 325 x 288

pixels.

Table 5: Description of the thermal videos.

Name Time Description

ip 9 Interior. A person sitting at a table.

ip2 24

Interior. A person sitting at a table, remains a

few seconds and then it rises.

iev 51

Interior. A person crosses a corridor lit with

fluorescent at different times.

ief 38

Interior. Two people stand before food

machines. Extra people appear and disappear

in the scene.

ep 29

Exterior. The scene is empty much of the time,

except for one person that appears, remains

static for a few seconds and then leaves.

eev 29

Exterior. Scene empty most of the time, except

for one person who crosses a moderate speed.

eef 29

Exterior. A few people move to the back of the

stage. Soon a person is in the foreground, is

fixed a few seconds and then leaves.

eef2 29

Exterior. The scene is empty much of the time,

except for one person that appears, remains

static for a few seconds and then leaves.

3 DEVELOPMENT

OF THE ALGORITHM

So then he decided to create a tool prior calibration.

Figure 3: Functionality of the detector algorithm.

BACKGROUNG

GENERATION

PRE-DETECTION

THRESHOLDING ELIMINATION SMOOTHING

FINALISATION

SEGMENTATION

Input frames

Parameters of detection

DetectionandTrackingoftheHumanHotSpot

327

With this tool, the user designates (indirectly) the

temperature ranges to be searched. Each user must

analyze for your particular case scenario, choose as

representative situation possible and take a thermal

snapshot. With this calibration image, the user can

develop a set of parameters that the detection

algorithm used for detection. It is also possible (if

the particular case requires) designate areas of no

interest , regions where the algorithm simply discard

the information collected.

With this tool calibration algorithm was

practically finished and adopted the following

structure, identical to the final version. The

description of the subsystems is the following;

 Background generation: generates the

background of the current detection information

from the captured frame and the funds generated

in previous detections.

 Segmentation: using generated parameters

calibration algorithm to remove information

considered irrelevant.

 Pre-detection: subtracts the background

generated segmented frame and passes the result

to grayscale.

 Thresholding: set an even clearer difference

between lights and background.

 Elimination: performs a clean image, eliminating

small and unimportant areas.

 Smoothing: recovers lost data or reconstructed in

previous steps.

 Finalisation: classifies heat sources and found a

box marked with different color depending on

the type.

(a) Acceptable detection (b) Improving detection

Figure 4: Detection Effectiveness.

4 RESULTS

Detection is considered "acceptable" if the algorithm

has been able to frame a person properly classify its

evolution, while detection is called "improved" if the

person framing elements are included scenario, if the

figure of a person has been fragmented into more of

a focus, or if its evolution has been misclassified.

Figure 5: Detection of “ip”, frames 99-118.

Some examples for some of the different types of

data recorded in Figures 5-8 are displayed.

Applying these criteria resulted in the videos,

and the parameters of the our dataset the following

table 6 is obtained.

For each video total detections in which people

appear in the second column is specified. In the third

column these detections are distinguished acceptable

in the upper part and the bottom improvable. In the

fourth column the percentage being relative to the

total detections people.

Figure 6: Detection of “eef2”, frames 390-409.

Figure 7: Detection of “ief”, frames 99-118.

BIOSIGNALS2014-InternationalConferenceonBio-inspiredSystemsandSignalProcessing

328

Figure 8: Detection of “iev”, frames 158-177.

It is noted that these criteria, the algorithm does not

seem particularly effective, but if, for example, is

considered an acceptable detection to detect a person

at different points, only for IP and IP2 videos would

be 100% of detections.

Table 6: Accuracy on the person detection.

Type of

video

Total

Acceptable /

Improvable

Accuracy (%)

IP 33

20 60,61

13 39,39

IP2 113

42 37,17

71 62,83

IEV 86

38 44,18

48 55,81

IEF 121

35 28,93

86 71,07

EP 45

8 17,78

37 82,22

EEV 12

12 100

0 0

EEF 73

2 2,74

71 97,26

EEF2 85

8 9,41

77 90,59

5 DISCUSSION

AND CONCLUSSIONS

This work has developed an algorithm detector heat

sources. This algorithm needs the support of a

calibration tool , which is a graphical user interface

that generates some of the sensing parameters. To do

this the user must use a fixed thermal image (photo),

representative of the scenario in question. The

sensing device consists of an infrared camera with

its video output connected to the input of an

acquisition card inserted into a computer. Recording

in AVI format generated is input to the algorithm,

which generates a video output with the same

number of frames, and frames per second that the

detected foci are framed by a rectangle. These boxes

can be different colors and indicate the nature of the

sources found. By default, the cuasidinámicos

framed by orange lights, that is, those who are

continuously detected in the scene and whose

position and dimensions vary slightly over time.

They are framed in blue static red lights and

dynamic . When a scene foci appear and disappear

continuously, are considered static those whose

position and dimensions vary slightly detection in

dynamic detection and those whose evolution is

more noticeable.

ACKNOWLEDGEMENTS

This work is supported by funds from The Spanish

Government, under Grant MCINN TEC2012-38630-

C04-02.

REFERENCES

Liew, A.L., Lim, A. Kwoh, L., 2005. A Stochastic Model

for Active Fire Detection Using the Thermal Bands of

MODIS Data. In IEEE Geoscience and Remote

Sensing Letters, Vol. 2, Issue 3, pp. 337 – 341.

Vodacek, A., Kremens, R.L., Ononye, A., Tang, C., 2005.

A Hybrid Contextual Approach to Wildland Fire

Detection Using Multispectral Imagery, In IEEE

Geoscience and Remote Sensing Letters, 43(9), pp.

2115 - 2126.

Ishino, R., 2002. Detection of a Faulty Power Distribution

Apparatus by Using Thermal Images. In IEEE Power

Engineering Society Winter Meeting, Vol. 2, pp. 1332

– 1337.

Carter, L.J., O’Sullivan, M.J., Hung, Y.J., Teng, J.C-C.,

1998. Thermal Imaging for Landmine Detection. In

Second International Conference on the Detection of

Abandoned Land Mines, pp. 110 – 114.

Azak, M.D., Akgün, S., Azak, S.I., Torun E., 2003.

Thermal Detection of Buried Circular Objects with a

Rule-Based Fast Shape Detection Algorithm. In

Proceedings of IEEE Sensors, Vol.2, pp. 765 - 768.

Heo, J., Savvides, M., Vijayakumar, B.V.K., 2005.

Performance Evaluation of Face Recognition using

Visual and Thermal Imagery with Advanced

Correlation Filters. In IEEE Computer Society

Conference on Computer Vision and Pattern

Recognitions, pp. 9.

Socolinsky, D.A., Selinger, A., 2004. Thermal Face

Recognition in an Operational Scenario. In

Proceedings of the IEEE Computer Society

Conference on Computer Vision and Pattern

Recognition, pp. 1012 - 1019.

Jiang, L., Yeo, A., Nursalim, J., Wu, S., Jiang, X., Lu,

DetectionandTrackingoftheHumanHotSpot

329

Z., 2004. Frontal Infrared Human Face Detection by

Distance from Centroid Method. In Proceedings of

International Symposium on Intelligent Multimedia,

Video and Speech Processing, pp. 41 - 44.

Jiang, G., Song, X., Zheng, F., Wang, P., Omer A.M.,

2005. Facial Expression Recognition Using Thermal

Image. In 27th Annual International Conference on

the Engineering in Medicine and Biology Society, pp.

631 - 633.

Iwasawa, S., Ebinara, K., Morishima, J.O., Shigeo, 1998.

Real-Time Human Posture Estimation using

Monocular Thermal Images. In 3rd IEEE

International Conference on Automatic Face and

Gesture Recognition, pp. 492 - 497.

Imai, M., Takayuki, N., Shida, T., Sato, M., Ito, R.,

Akamine, I., 1993. Thermal Image Proccesing using

Neural Network. In IJCNN’93, Proceedings of

International Joint Conference on Neural Networks,

Vol. 3, pp. 2065 - 2068.

Lin, C.L., Fan, K.C. 2004. Biometric Verification using

Thermal Images of Palm-Dorsa Vein Patterns. In IEEE

Transactions on Circuits and Systems for Video

Technology, Vol. 14, Issue 2, pp. 199 - 213.

Cromer, A.H., 1985. Física para las ciencias de la vida.

Ed. Reverté. , 2

edition.

BIOSIGNALS2014-InternationalConferenceonBio-inspiredSystemsandSignalProcessing

330