Volume 8, Issue 5, May – 2023 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165

Systematic Literature Review of Pedestrian Detection

using the YOLO Algorithm
Lamsadi1 *, Arief Setyanto2, Tonny Hidayat3
1
Informatics Engineering Study, Amikom University of Yogyakarta
2
Informatics Engineering Study, Amikom University of Yogyakarta
3
Informatics Engineering Study, Amikom University of Yogyakarta

Abstract:- Technology is developing so rapidly at this in the field of object detection. Object detection is the
time. Every time various latest and cutting-edge lifeblood of Computer Vision and Image Processing [1].
technologies in various fields transmit life. One of them There are 4 main focuses in Computer Vision, namely
is in the field of object detection. As technology recognition, visual tracking, semantic segmentation and
develops, the need for object detection systems becomes image restoration [2]. To be able to count objects
very strong. Object detection or object detection is the automatically, the first two things that must be done are to
lifeblood of Computer Vision and Image Processing. detect and classify objects (movable or immovable), for
There are 4 main focuses in Computer Vision, namely example vehicles, pedestrians and others [20]. You Only
Recognition, Visual Tracking (visual tracking), Live Once or better known as YOLO is a very well-known
Semantic Segmentation (semantic segmentation) and and widely used algorithm. YOLO is a specific algorithm
Image Restoration (image restoration). To be able to do for detecting objects [10]. In recent years, the YOLO
these four things, we need an algorithm that can Algorithm has shown interesting results in various areas in
effectively be applied to detect objects, especially object detection[12], both large-scale and special, has
pedestrians, so YOLO was chosen as the answer. YOLO solved many problems in the field of object detection in
is one of several algorithms that are often used in general, vehicle registration plate detection, pedestrians and
Machine Learning. You Only Live Once or better others [9]. Even now YOLO (Object detection) can be used
known as YOLO is a very well-known and widely used to control production processes in factories based on video
algorithm. YOLO is a specific algorithm for object data in real time [16]. In several studies YOLO is not only
detection. In recent years, the YOLO Algorithm has used to detect objects in the form of humans. Yolo can also
shown interesting results in various areas of object be used to detect fish movements in the water [18].
detection, both large-scale and special, has solved many
problems in the field of object detection in general, the II. RESEARCH METHODS
detection of license plates of vehicles, pedestrians, etc.
Through this systematic literature review, it is hoped SLR is a method or approach in research, namely by
that it will be able to provide enlightenment for the reviewing some of the literature by reviewing a particular
development of Object Detection science. topic and emphasizing a focused question. Then the
questions are selected, identified, assessed and concluded
Keywords:- Object Detection; Image Processing; YOLO; with predetermined criteria. In addition, SLR also aims to
Pedestrian; Machine Learning; Systematic Literature find research gaps so that new research areas emerge that
Review. have the opportunity to be studied.

I. INTRODUCTION A. Research Questions

To conduct SLR research, several steps or criteria are
Technology is developing so rapidly at this time. needed, called PICOC. PICOC stands for Population,
Every time, various new and cutting-edge technologies Intervention, Comparison, Outcomes and Context. The
emerge in various fields and aspects of life. One of them is following is a PICOC summary table.

Population Image processing, machine learning

Intervention Datasets, Models and Methods
Comparasion Accuracy, precision, recall, performance and speed
Outcomes Model detection accuracy and performance efficiency
Context Pedestrian and YOLO
Table 1:- PICOC Summary

To complete research using the SLR method, several questions or Research Questions (RQ) are needed as below so that the
research is more focused, directed and conical.

IJISRT23MAY1184 www.ijisrt.com 1420

Volume 8, Issue 5, May – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
Table 2:- Research questions
ID Research Questions Purpose
RQ1 What is the research on object detection especially on Identification of research developments on object
pedestrians like in the last 6 years? detection, especially on pedestrians.

RQ2 Where does the research source used as a reference for Identification of research sources as references.
object detection especially pedestrians using YOLO
come from?
RQ3 From which countries did the researchers raise Identify countries that have done a lot of research on
research on object detection, especially on object detection, especially on pedestrians.
pedestrians?
RQ4 What and how is the object called the pedestrian? Identify the pedestrian concept

RQ5 What and how do pedestrian objects becomes vital Identify pedestrian safety systems
objects?
RQ6 What and how does the YOLO Algorithm work? Identify the concepts and how the Yolo Algorithm
works.

B. Study Selection III. RESULTS AND DISCUSSION

In conducting study selection, the research included in
this SLR is research published within the last 6 years in the The following is a discussion of the literature that was
form of journals or conferences. There are 3 main collected.
keywords used in this study, namely object detection,
pedestrian and YOLO. In general, there are two categories A. Research Years
of research taken as literature, namely experimental and In the last 6 years, research in the field of object
survey research. The stages of the literature search process detection, especially those using the YOLO algorithm,
to find the right one are shown in Figure 1. seems to have progressed quite rapidly, although there has
been a slight decline in 2021, this does not mean that this
research is less interesting. The most research is in 2022
and this indicates that as more years are added it is likely to
increase. The following is a graphical image of the
distribution with the percentage of research years.

Fig 2:- Research Years

B. Research Sources
The databases used for research were ScienceDirect
(sciencedirect.com), IEEE Xplore (ieeexplore.ieee.org),
Springer (springerlink.com), IET Search
Fig 1:- Research Search Flowchart (ietresearch.onlinelibrary.wiley.com), SPIE (https://www.
spiedigitallibrary.org ), MDPI (https://www.mdpi.com) and
Software Impact (www.softwareimpacts.com). Details of
the distribution of sources are explained in Figure 3 below:

IJISRT23MAY1184 www.ijisrt.com 1421

Volume 8, Issue 5, May – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
In general, pedestrians are divided into 2 (two),
namely real or real people (actual person) and pseudo
people (the depiction of a person). Depictions of people can
be in the form of pictures of people, statues or dolls and so
on [3]. In other cases pedestrians are also called Passengers
(Passengers). This can be seen from the location where they
are located, for example at a bus terminal or train station
(Sipetas et al., 2020). Detection of the movement of objects
(people) or the Euclidian distance between objects and their
surroundings can be important information in
distinguishing between real and imaginary people [24].

Even though there are lots of traffic indicators,

Fig 3:- Research Sources crossroads and pedestrian safety signs, the possibility of
accidents between vehicles (cars, motorcycles, trains and
As can be seen in the ScienceDirect graph, it occupies others) and pedestrians is still very high. Therefore, the
the majority or the main choice as a research source. This is development of advanced cognitive systems (eg, pedestrian
due to ScienceDirect's excellent reputation as a provider of detection) is a promising step towards a rapid reduction in
world-class journals. the number of traffic accidents. Recently, the development
of a pedestrian safety system has received a lot of positive
C. Countries response and attention. However, pedestrian safety is quite
Countries that are so intensively conducting research a difficult job due to reasons such as illumination and
in the field of object detection, especially with the Yolo appearance effects (texture, ratio, area)[8]. Therefore we
Algorithm, are still dominated by two countries, namely need a system that can distinguish between pedestrians and
China and Egypt, even though the percentage distance the environment, such as ITMSs and SMRIK. ITMSs or
between these two countries is quite far. This is as shown in Intelligent Traffic Management Systems is a system that
Figure 4 below: not only uses object detection and image processing
approaches but also Geometric Computing [21]. SMRIK is
a Pedestrian Automatic Emergency Braking (PAEB)-based
machine learning [28].

E. Yolo Algorithm
The YOLO algorithm is a method based on regression
that predicts through bounding boxes and class objects to
determine the location of an object in an image using a
Single Neural Network. (Feng et al., 2019; Yu, J., & Choi,
H. 2022; Saada et al., 2022). The YOLO algorithm works
by dividing the image (image) into several parts (cells),
each cell is used to predict a number of bounding boxes if
there is more than one object in the image. (Feng et al.,
2019; Xue et al., 2021; Han et al., 2021). Then the
prediction results will be collected and the bounding box
with the smallest probability will be removed. The
bounding box with the largest predicted probability value
will be the final result. (Han et al., 2021). The following is
Fig 4:- Countries an example of the bounding box and class object shown in
Figure .5:
D. Object Types
Research on object detection, especially pedestrians,
is very important. This is due to the increasing number of
devices or technologies that use this system. An example is
autonomous vehicles, which are increasingly developing
day by day because more and more are doing research [29].
Not only that, integration between object detection and
other branches of science can support the health of
pedestrians with air control systems on the road [30].
Pedestrian detection is the basis of many human-based
tasks, including speed tracking, detection of pedestrian
movement, automation of pedestrian recognition and
Fig 5. bounding box and class object
appropriate response actions or reject detection of pseudo
Source (Han et al., Procedia Computer Science, volume :
pedestrians [8].
183, page : 63 https://doi.org/10.1016/j.procs.2021.02.031)

IJISRT23MAY1184 www.ijisrt.com 1422

Volume 8, Issue 5, May – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
Yolo is capable of detecting both 2D and 3D objects Machine Learning with Application, 6, 10013.
in [27] the form of images (static images) or videos [19]. https://doi.org/10.1016/j.mlwa.2021.100134 , 2021.
Yolo is able to detect objects in real time[5]. At a time [3]. Donoso, F.G., Amoros, J.C., Escalona, F., & Cazorla,
when COVID-19 cases were still rampant, Yolo was used M, “Three-dimensional reconstruction using SFM for
to detect social distancing and [22] the use of face masks actual pedestrian classification”, Expert Systems With
and was found to have an accuracy rate of up to 90% in 150 Applications, 213, 119006.
facial samples tested [7]. Yolo can process images up to 45 https://doi.org/10.1016/j.eswa.2022.119006, 2022.
frames per second (FPS) [4]. [4]. Feng, X., Jiang, Y., Yang, X., Du, M., & Xin Li, X,
“Computer vision algorithms and hardware
Several years ago, experts have published several implementations: A survey”, Integration, the VLSI
YOLO versions such as YOLO V2, YOLO V3, YOLO V4, Journal, 69: 309–320.
YOLO V5 and There are several limited-revision versions, https://doi.org/10.1016/j.vlsi.2019.07.005 China,
such as YOLO-LITE [10]. Yolov3 is capable of accurately 2019.
detecting up to 79% of pedestrians out of 20,000 detected [5]. Hana, Z., Huanga, H., Fan, Q., Li, Y., Li, Y., & Chen,
objects [11]. The latest generation of the YOLO Algorithm X, “SMD-YOLO: An Efficient And Light Weight
is YOLOv5 and uses the Python programming language Detection Method For Mask Wearing Status During
unlike its predecessors which still use the C [28][13]. The COVID-19 Pandemic”, Computer Methods and
However, in terms of access speed and detection with an Programs in Biomedicine, 221, 106888.
accuracy that is not inferior to YOLO in general, Tiny Yolo https://doi.org/10.1016/j.cmpb.2022.106888, 2020.
still dominates [26]. [6]. Han, X., Chang, J., & Wang, K, “Real-Time Object
Detection Based On YOLO-V2 For Tiny Vehicle
IV. CONCLUSION Object”, 10th International Conference of Information
and Communication Technology (ICICT-2020),
This SLR research aims to identify and analyze Procedia Computer Science, 183: 61–72.
pedestrian types based on the nature of the object and its https://doi.org/10.1016/j.procs.2021.02.031, 2022.
location. In general, pedestrians are divided into 2 (two), [7]. Hou, Y. C., Baharuddin, M. Z., Yussof, S. &
namely real or real people (actual person) and pseudo Dzulkifly, S, “Social Distancing Detection with Deep
people (the depiction of a person). Pedestrians are also Learning Model”, 2020 8th International Conference
called passengers. By identifying and analyzing on Information Technology and Multimedia (ICIMU):
pedestrians, they will be able to maximize the performance 334-338.
of the YOLO Algorithm. The YOLO algorithm is a method https://doi.org/10.1109/ICIMU49871.2020.9243478,
based on regression that predicts through bounding boxes 2020.
and class objects to determine the location of an object in [8]. Iftikhar, S., Asim, M., Zhang, Z., & El-Latif, A. A. A,
an image using a Single Neural Network. Of the 33 “Advance Generalization Technique Through 3D
literatures reviewed, 78% or the majority of research CNN To Overcome The False Positives Pedestrian In
sources were from sciencedirect.com. China is the country Autonomous Vehicles”, Telecommunication Systems,
with the highest number of studies, reaching 37%, followed volume 80: pages 545–557.
by India, 12% and the others below 10%. The year of https://doi.org/10.1007/s11235-022-00930-1, 2020.
research from the literature reviewed came from 2022 as [9]. Jamtsho, Y., Riyamongkol, P., & Waranusast, R,
much as 40% and 2020 as much as 27%. “Real-Time Bhutanese License Plate Localization
Using YOLO”, The Korean Institute of
In the end object detection based on the YOLO Communications and Information Sciences (KICS),
algorithm becomes very important and needed when it ICT Express, 6: 121–124.
comes to the safety of pedestrians, especially on roads or https://doi.org/10.1016/j.icte.2019.11.001, 2020.
around public transportation roads such as trains. The [10]. Jiang, P., Ergu, D., Liu, F., Cai, Y., & Ma, B, “A
implementation of an automatic emergency braking system Review of Yolo Algorithm Developments”, The 8th
such as SMRIK provides an additional sense of security for International Conference on Information Technology
pedestrians [28]. and Quantitative Managemen, Procedia Computer
Science, 199: 1066–1073.
REFERENCES https://doi.org/10.1016/j.procs.2022.01.135, 2022.
[11]. Kataev, G., Varkentin, V., & Nikolskaia, K, “Method
[1]. Arulprakash, E., & Aruldoss, M, “A Study On To Estimate Pedestrian Traffic Using Convolutional
Generic Object Detection With Emphasis On Future Neural Network”, XIV International Conference 2020
Research Directions”, Journal of King Saud SPbGASU “Organization and safety of traffic in large
University – Computer and Information Sciences, cities”, Transportation Research Procedia, 50: 234–
34(9):7347– 241. https://doi.org/10.1016/j.trpro.2020.10.029,
7365,https://doi.org/10.1016/j.jksuci.2021.08.001, 2020.
2022. [12]. Kumar, A., Kalia, A., & Kalia, A, “ETL-YOLO v4: A
[2]. Chai, J., Zeng, H., Li, A., & Ngai, Eric, E.W.T, “Deep Face Mask Detection Algorithm In Era Of COVID-19
Learning in Computer Vision: A Critical Review of Pandemic”, Optik - International Journal for Light
Emerging Techniques and Application Scenarios”,

IJISRT23MAY1184 www.ijisrt.com 1423

Volume 8, Issue 5, May – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
and Electron Optics, 259, 169051. [23]. Saada, M., Kouppas, C., Li, B., & Meng, Q, “A
https://doi.org/10.1016/j.ijleo.2022.169051, 2022. Multi-Object Tracker Using Dynamic Bayesian
[13]. Li, P., & Zhao, W, “Image Fire Detection Algorithms Networks And A Residual Neural Network Based
Based On Convolutional Neural Networks”, Case Similarity Estimator”, Computer Vision and Image
Studies in Thermal Engineering, 19, 100625. Understanding, 225, 103569.
https://doi.org/10.1016/j.csite.2020.100625, 2020. https://doi.org/10.1016/j.cviu.2022.103569, 2022.
[14]. Li, S., Li, Y., Li, Y., Li, M., & Xu, X, “YOLO-FIRI [24]. Sipetas, C., Keklikoglou, A., & Gonzales, E. J,
: Improved YOLOv5 for Infrared Image Object “Estimation Of Left Behind Subway Passengers
Detection”, IEEE Access, VOLUME 9: 141861- Through Archived Dataand Video Image Processing”,
141875. Transportation Research Part C, 118, 102727.
https://doi.org/10.1109/ACCESS.2021.3120870, https://doi.org/10.1016/j.trc.2020.102727, 2020.
2021. [25]. Shinde, S., Kothari, A., & Gupta, V, “YOLO based
[15]. Loey, M., Manogaran, G., Taha, M. H. N., & Khalifa, Human Action Recognition and Localization”,
N. E. M, “Fighting against COVID-19: A Novel International Conference on Robotics and Smart
Deep Learning Model Based On Yolo-V2 With Manufacturing (RoSMa2018)”, Procedia Computer
Resnet-50 For Medical Face Mask Detection”, Science, 133, 831–838.
Sustainable Cities and Society, 65, 102600. https://doi.org/10.1016/j.procs.2018.07.112, 2018.
https://doi.org/10.1016/j.scs.2020.102600, 2021. [26]. Shuo, Z., Yanxia, W., Chaoguang, M., & Xiaosong,
[16]. Malburg, L., Rieder, M. P., Seiger, R., Klein, P., & L, “Tiny YOLO Optimization Oriented Bus Passenger
Bergmann, R, “Object Detection for Smart Factory Object Detection”, Chinese Journal of Electronic,
Processes by Machine Learning”, The 4th 29(1): 132-138.
International Conference on Emerging Data and https://doi.org/10.1049/cje.2019.11.002 , 2020.
Industry 4.0 (EDI40) March 23 - 26, 2021, Warsaw, [27]. Simon, M., Amende, K., Kraus, A., Honer, J.,
Poland, Procedia Computer Science, 184: 581–588. Samann, T., Kaulbersch, T., & Milz, S, “Complexer-
https://doi.org/10.1016/j.procs.2021.04.009, 2021. YOLO: Real-Time 3D Object Detection and Tracking
[17]. Marcos, A. N., Gorka Azkune, G., & Carreras, I. A, on Semantic Point Clouds”, Proceedings of the
“Egocentric Vision-based Action Recognition: A IEEE/CVF Conference on Computer Vision and
survey”, Neurocomputing, 472 (2022), 175–197. Pattern Recognition (CVPR) Workshops 2019, pp. 0-
https://doi.org/10.1016/j.neucom.2021.11.081, 2022. 0. https://openaccess.thecvf.com, 2019.
[18]. Mohamed, H. E., Fadl, A., Anas, O., Wageeh, Y., El [28]. Socha, K., Borg, M., & Henriksson, J, “SMIRK: A
Masry, N., Nabilm, A., & Atia, A, “MSR-YOLO: Machine Learning-Based Pedestrian Automatic
Method to Enhance Fish Detection and Tracking in Emergency Braking System With A Complete Safety
Fish Farms”, The 11th International Conference on Case”, Software Impacts, 13, 100352.
Ambient Systems, Networks and Technologies (ANT), https://doi.org/10.1016/j.simpa.2022.100352, 2022.
Warsaw, Poland, Procedia Computer Science, 170: [29]. Song, X., Gao, S., & Chen, C, “A Multispectral
539–546. https://doi.org/10.1016/j.procs.2020.03.123, Feature Fusion Network For Robust Pedestrian
April 6 - 9, 2020 Detection”, Alexandria Engineering Journal, 60: 73–
[19]. Molchanov, V. V., Vishnyakov, B. V., Vizilter, Y. 85. https://doi.org/10.1016/j.aej.2020.05.035, 2021.
V., Vishnyakova, O. V., & Knyaz, V. V, “Pedestrian [30]. Verstaevel, N., Barthélemy, J., Forehead, H., Arshad,
Detection In Video Surveillance Using Fully B., & Perez, P, “Assessing The Effects Of Mobility
Convolutional YOLO Neural Network”, Proceedings On Air Quality: The Liverpool Smart Pedestrian
Volume 10334, Automated Visual Inspection and Project”, World Conference on Transport Research –
Machine Vision II , 103340Q. WCTR 2019, Mumbai, 26-30 May 2019,
https://doi.org/10.1117/12.2270326 , 2017. Transportation Research Procedia, 48: 2197–2206.
[20]. Namatevs, I., Sudars, K., & Polaka, I, “Automatic https://doi.org/10.1016/j.trpro.2020.08.276, 2020.
data labeling by neural networks for the counting of [31]. Wu, P., Li, H., Zeng, N., & Li, F, “FMD-Yolo: An
objects in videos”, Procedia Computer Science, 149: Efficient Face Mask Detection Method For COVID-
151–158. https://doi.org/10.1016/j.procs.2019.01.118, 19 Prevention And Control In Public”, Image and
2019. Vision Computing, 117, 104341.
[21]. Namazi, E., Mester, R., Lu, C., & Li, J, “Geolocation https://doi.org/10.1016/j.imavis.2021.104341, 2022.
Estimation Of Target Vehicles Using Image [32]. Xue, L., Yan, W., Luo, P., Zhang, X., Chaikovska, T.,
Processing and Geometric Computation”, Liu, K., Gao, W., & Yan, K, “Detection and
Neurocomputing, 499: 35–46. localization of hand fractures based on GA_Faster R-
https://doi.org/10.1016/j.neucom.2021.10.127, 2022. CNN”, Alexandria Engineering Journal, 60: 4555–
[22]. Prasad, J., Jain, A., Velho, D., & K S, S.K, “COVID 4562. https://doi.org/10.1016/j.aej.2021.03.005, 2021.
Vision: An Integrated Face Mask Detector And Social [33]. Yu, J., & Choi, H, “YOLO MDE: Object Detection
Distancing Tracker”, KeAi Chinese Roots Global with Monocular Depth Estimation”, MDPI
Impact International Journal of Cognitive Computing Electronics, 11(1), 76.
in Engineering, 3, 106–113. https://www.doi.org/10.3390/electronics11010076,
https://doi.org/10.1016/j.ijcce.2022.05.001, 2022. 2022.

IJISRT23MAY1184 www.ijisrt.com 1424