International Journal of Engineering and Science Applications

ISSN 2406-9833
IJEScA

Autonomous Quadcopter Based Human Detection System

for Search and Rescue
F G Al Ayubi1, Irmian1, R Uswarman1, S Istiqphara1
1
Electrical Engineering, Institut Teknologi Sumatera (ITERA)
Jl Terusan Ryacudu, 35365, Lampung Selatan, Indonesia
E-mail: [email protected]

ABSTRACT

The development of navigation and tracking of an object using unmanned aerial vehicles (UAV) has been
evolved more advanced. The utilization of this technology is scattered throughout the area including for
search and rescue in small and large areas. In this research, the UAV used is a multirotor based on camera
for human detection that has been developed and integrated into a fully autonomous system. The first
stage is designing a controller and the second step is to integrate the companion computer on the
quadcopter that has the OpenCV library installed with the HOG and SVM methods. The system presented
in this paper is able to perform the stable flying via PID controller. The output response of roll, pitch, and
yaw has an overshoot of 0.5 %. Identify target in the form of human body in real-time and identify target
location based on the direction of the Earth's compass with test result has average error, latitude 1 =
0.026%, longitude 1 = 0.034%, latitude 2 = 0.016%, and longitude 2 = 0.00168%.

Keywords: UAV; multirotor; HOG; SVM; PID

1. INTRODUCTION when hovering using Proportional Integral

Unmanned Aerial Vehicle (UAV) with an Derivative (PID) that we have designed. Then in
autonomous system developed rapidly in the last the second phase, it performs human detection
few years [1] including search and rescue purpose programming directly on the companion computer
[2]. Search and rescue in disaster areas with via the Histogram of Oriented (HOG) method and
extensive terrain or on locations with dangerous Support Vector Machine (SVM). In this research,
conditions that could threaten the rescue team. In quadcopter was built with a Pixhawk flight
this research, the UAV technology used is a controller, odroid-xu4 as a companion computer
quadcopter that can perform vertical take-off and to detect human objects, four BLDC motorcycles
landing (VTOL). In the process of search and as well as a propeller, and used LiPo batteries as a
rescue for disaster areas required autonomous source of the voltage.
quadcopter with knowing the real-time location
and the position of starting point when flying. 2. METODOLOGY
Previous research related to object detection has A. Autonomous System
been conducted [3] [4] [5]. The first phase of the A companion computer system is required
study was conducted in the design of a quadcopter to control the quadcopter such as VTOL
control system aimed at stabilizing the quadcopter autonomous. The companion computer system

60
Odroid XU4 needed to control the Pixhawk in parameters of PID use S-shape reaction curve
order to perform actions such as auto vertical where consist of time constant 𝑇 and delay time 𝐿.
takeoff, altitude hold for 30 seconds, to the
specified waypoint and autoland. The serial for
the UART driver module is required for Odroid
communication XU4 Pixhawk through MavLink
communication.

Fig. 3. S-shape reaction curve

The built-in PID control system will be used to set

the pitch, roll, and yaw of the quadcopter. Based
on graphic analyze, we get 𝐿 and 𝑇 then PID
parameter will be obtained using Table 1.
Table 1. Tuning PID via Ziegler Nichols type 1
Controller Type Kp Ti Td
P T/L - 0
PI 0,9 T/L L/0,3 0
PID 1,2 T/L 2L 0,5L

Fig. 1. Diagram of autonomous quadcopter C. Histogram of Oriented Gradients (HOG)

HOG is an image processing method used
B. PID Controller to detect objects (humans). This method calculate
The PID controller is a feedback gradients within a given area in an image. The
mechanism controller that is commonly used in object detection process is performed per frame of
industrial control systems. The PID controller the image captured as a result of video counting.
consists of proportional, integral, and derivative. The video counting frame has different
The block diagram of PID method can be seen in characteristics shown through the gradient
Fig. 1. distribution. This characteristic is implemented by
dividing the image into small windows (cells).
Each of these cells will be merged back to a
histogram of a gradient. The combination of this
histogram will form the characteristic is the
implementation of the image representing the
Fig. 2. Block control Diagram PID object. The following extraction process of the
HOG method is done which will be described in
The parameters of the PID are obtained
Fig. 4 [6].
using the Ziegler Nichols type 1. Tuning

61
 Weight Vote into
Normalize Gamma
Input Image Compute Gradients Spatial &
& Colour
Orientation Cell

Contrast Normlize
Person / non-person Colect HOG’s Over
Linier SVM over Overlapping
Classification Detection Window
Spatial Block
Fig. 4. HOG Method Extraction Process

D. Support Vector Machine (SVM) Fig. 6. Quadcopter Frame

The Support Vector Machine (SVM) is a
method used to identify and distinguish humans or The image above is a representation of the

non-humans. The process of this method works B-Frame and E-Frame that can determine the

when the HOG framework on OpenCV video linear position of the Quadcopter (r E).
𝑟 𝐸 = [𝑋 𝑌 𝑍]𝑇 (1)
counting results from video captured by webcam.
Θ𝐸 = [𝜙 𝜃 𝜓 ]𝑇 (2)
In this process, SVM serves as a separator of two
When flying, quadcopter will generate a
classes in input space between -1 and + 1. Class-
rotation matrix is a B-Frame transformation into
1 is symbolized by a red box and class + 1 is
E-Frame use the matrix transfer 𝑇𝜃 :
symbolized by a yellow circle. The basic principle
𝑝 𝜙
of SVM itself is as a linear classifier that can be [𝑞 ] = 𝑇𝜃−1 [ 𝜃 ], (3)
seen in Fig. 5 [7]. 𝑟 𝜓
where the transformation matrix from B-Frame to
E-Frame:
1 0 −𝑠𝜃
𝑇𝜃−1 = [0 𝑐𝜙 𝑐𝜃 𝑠𝜙 ] (4)
0 −𝑠𝜙 𝑐𝜙 𝑐𝜃

1 −𝑠𝜙 𝑡𝜃 𝑠𝜃 𝑡𝜃
Fig. 5. The SVM process finds the best 0 𝑐𝜙 −𝑠𝜙
𝑇𝜃−1 = 𝑠𝜙 𝑐𝜙 (5)
hyperplane to separate both class-1 and + 1 0
[ 𝑐𝜃 𝑐𝜃 ]
E. Mathematical Model of Quadcopter The equation (6) – (11) was obtained a
The mathematical modelling of general quadcopter equation derived from the
quadcopter essentially refers to the quadcopter Euler-Newton method
motion or can be called degrees of freedom. The 𝑈1
𝑋̈ = (𝑠𝑖𝑛𝜓𝑠𝑖𝑛𝜙 + 𝑐𝑜𝑠𝜓𝑠𝑖𝑛𝜃𝑐𝑜𝑠𝜙) (6)
𝑚
quadcopter has 6 Degree of Freedom (DoF) which
𝑈1
𝑌̈ = (−𝑐𝑜𝑠𝜓𝑠𝑖𝑛𝜙 + 𝑠𝑖𝑛𝜓𝑠𝑖𝑛𝜃𝑐𝑜𝑠𝜙) (7)
is divided into two reference frames i.e. E-Frame 𝑚
is a fixed axis that is earth and B-Frame is a 𝑈1
𝑍̈ = −𝑔 + (𝑐𝑜𝑠𝜃𝑐𝑜𝑠𝜙) (8)
𝑚
quadcopter moving axis that can be seen in Fig. 6 𝐼𝑌𝑌 − 𝐼𝑍𝑍 𝐽𝑇𝑃 𝑈2
𝑝̇ = 𝑞𝑟 − 𝑞Ω + (9)
[8] [9]. 𝐼𝑋𝑋 𝐼𝑋𝑋 𝐼𝑋𝑋
𝐼𝑍𝑍 −𝐼𝑋𝑋 𝐽 𝑈3
𝑞̇ = 𝐼𝑌𝑌
𝑝𝑟 − 𝐼 𝑇𝑃 𝑝Ω + 𝐼𝑌𝑌
(10)
𝑌𝑌
𝐼𝑋𝑋 −𝐼𝑌𝑌 𝑈4
𝑟̇ = 𝐼𝑍𝑍
𝑝𝑟 − 𝐼𝑍𝑍𝑍
(11)

62
 While the input of the general equation
quadcopter (6)-(11) is determined from the
equation (12)-(16) which is the velocity of the
propeller when add disturbance.
𝑈1 = 𝑏 (Ω12 + Ω22 + Ω23 + Ω24 ) (12)
𝑈2 = 𝑏𝑙 (−Ω22 + Ω24 ) (13)
𝑈3 = 𝑏𝑙 (−Ω12 + Ω24 ) (14)
𝑈4 = 𝑑 (−Ω12 + Ω22 −Ω23 + Ω24 ) (15)
Ω = −Ω1 + Ω2 − Ω3 + Ω4 (16)
After the transfer function is acquired
Fig. 8. Complete Hardware of Quadcopter
from the general equation (6)-(11), that take part
of the pitch, roll and yaw to be controlled using Table 2. Physical measuring parameters
Physical measuring quadcopter
PID. The tuning of PID controller parameters
Name Value
(proportional Integral differential) is always based
LxF (m) 0,16
on the review of the regulated characteristics
LyF (m) 0,17
(plant). LxB (m) 0,16
LyB (m) 0,17
F. Physical parameters identification
Massa (kg) 1,82 kg
The moment of inertia calculation is
necessary to know the value of Ixx, Iyy and Izz. Parameter identification using experiment
In the moment of inertia required data method. The experimental method is a technique
specifications of the quadcopter physical of measuring characteristics of the quadcopter
parameters. The specification can be seen in Table plant using physical measurements moment of
2. inertia with trivial pendulum method. The trivial
pendulum method is chosen because its
implementation is very easy. Measurements are
performed in X, Y and Z axes.

Fig. 7. Physical measuring parameters

Fig. 9. Trivial Pendulum

63
 After that a moment of inertial calculation B. Implementation Auto Take-off and Landing
with the equation (17): The experiment was carried out by giving

𝑀. 𝑔. 𝑅2 𝑇𝑥,𝑦,𝑧
2 the auto take-off and landing command to odroid-
𝐼𝑥𝑥,𝑦𝑦,𝑧𝑧 = (17)
4. 𝜋 2 . 𝐿 xu4 which had been connected to the Pixhawk. In
Figure 3.5, there are three parts of quadcopter
Finally, the moment of inertia can be seen in Table condition. The first is the quadcopter position on
3. hover in 10 seconds, the second is the quadcopter
Table 3. Moment of inertia state while running the mission to the specified
Moment of inertia (Kg.m2) waypoint in 16 seconds while in the third part is
Ixx 0,0232
the quadcopter state when it is landing.
Iyy 0,0228
Izz 0,0298

3. RESULT AND DISCUSSION

A. Simulation of PID Controller
PID controller simulation using Simulink.
The PID parameter is obtained using Ziegler
Nichols type 1 shown in the Table 4. The output Fig. 11. Position of Quadcopter
response of controller can be seen in Fig. 10. Fig. The height of the quadcopter is detected
10 show that the output have overshoot roll 6 %, using a barometer sensor. Quadcopter altitude
pitch 5.5 %, and yaw 8 %. never reaches at 6 meters according to the order in
odroid-xu4. The measured average constant
height value is 5.87 meters. However, In this
implementation, the thrust issued a motor of 70%
so it takes time to point the set-point about 8
seconds.

C. Object Detection
The following test results from the
autonomous quadcopter program run with remote
Fig. 10. Response of PID controller control of the PuTTY and VNC Viewer software
on a PC. All results are displayed in Figure 3.6
Table 4. The value of PID with the parameter number of objects, altitude,
Name P I D Overshoot
Roll 1,05 6,562 0,042 6% coordinates, flight mode and local time. The
Pitch 1,05 6,562 0,042 5.5%
Yaw 1,44 14,4 0,036 8% results of the detection are taken based on four
directions of scanning (north, east, south and
west) at one coordinate point with latitude data =-
5.3599526 and longitude = 105.3103595. In this
64
coordinate point area, there is one sample 4. CONCLUSIONS
prepared to test the program that was created. It
The result of the implementation
can be seen in Figure 3.6 that detected 1 person in
hardware is quadcopter can detect the human
the scanning area to the east and west.
object and fly stable with the average height that
successfully performed is 5.78 meters from the
specified 6 meters set-point. The result of the
coordinate test is done with the test coordinate
value obtained the measured coordinate value of
the system with the average error value of each
latitude 1 = 0.026%, longitude 1 = 0.034%,
Fig. 12. Human detection east direction latitude 2 = 0.016%, and longitude 2 = 0.0168%.
From the value, each of these errors can be stored
system has a fairly high level of accuracy.

ACKNOWLEDGEMENT
Thanks to Institut Teknologi Sumatera
(ITERA), this research is funded by ITERA Smart

Fig. 13. Human detection west direction Mandiri Research.

Table 5. Result of measurement of coordinate

REFERENCES
value on Rill's condition
[1] Aljehani, M., Inoue, M., 2019. Performance
Evaluation of Multi-UAV System in Post-
Disaster Application: Validated by HITL
Simulator. Journal of IEEE; (7), 64386-
64400.
[2] Lee, W.K., Lee, K.W., 2017. Experimental
Table 5 is field test coordinates from point one and Operation of Drone Micro-SAR with
two which contain error values of measured Efficient Time-Varying Velocity
coordinate values. The error value is obtained by Compensation. IET Journals and Magazines;
changing each coordinate value to a time value (in (53), 682-683.
seconds) so that the error value can be calculated. [3] Subrata, R.H., Tarumasely, R., Dwianto, C.,
When seen in table 5, the error value is obtained 2017. Perancangan Pengendali PID untuk
nearly zero, so the level of accuracy is high on the gerakan pitch dan roll pada quadcopter.
system. Journal of JETri; 14(2), 1-6.
[4] Gupta, B., Chaube, A., Negi, A., Goel, U.,
2017. Study on Object Detection using Open

65
 CV – Python. Journal of Computer [8] Bresciani, T., 2018. Modelling, Identification
Application; 162(8), 17-21. and Control of a Quadrotor Helicopter. M.S.
[5] Arifin, F., Daniel, R.A., Widiyanto, D., 2014. Thesis, Lund University, Lund Scania
Autonomous Detection And Tracking Of An Sweden
Object Autonomously Using Ar.Drone
Quadcopter. Journal of Computer Science
and Information; 7(1), 11-17.
[6] Suryadi, K., Sikumbang, S., 2015. Human
Detection Menggunakan Metode Histogram
Of Oriented Gradients (Hog) Berbasis
Open_Cv. Journal of Pendidikan Teknik
Elektro; 4(2).
[7] Dalal, N., Triggs, B., 2005. Histograms of
oriented gradients for human detection.
Computer Society Conference on Computer
Vision and Pattern Recognition (CVPR'05).

66