Моніторинг навігації натовпу під час надзвичайних ситуацій
The paper considers the task of crowd navigation monitoring, which might be performed using various sensors and technologies, with surveillance cameras being the most commonly employed. These cameras provide a video stream that typically lacks supplementary information. Extracting additional data fr...
Saved in:
| Date: | 2024 |
|---|---|
| Main Authors: | , , |
| Format: | Article |
| Language: | English |
| Published: |
The National Technical University of Ukraine "Igor Sikorsky Kyiv Polytechnic Institute"
2024
|
| Subjects: | |
| Online Access: | https://journal.iasa.kpi.ua/article/view/302774 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Journal Title: | System research and information technologies |
| Download file: |  |
Institution
System research and information technologies| _version_ | 1866302961817223168 |
|---|---|
| author | Tymoshchuk, Oksana Tishkov, Maksym Bondarenko, Victor |
| author_facet | Tymoshchuk, Oksana Tishkov, Maksym Bondarenko, Victor |
| author_sort | Tymoshchuk, Oksana |
| baseUrl_str | http://journal.iasa.kpi.ua/oai |
| collection | OJS |
| datestamp_date | 2025-02-09T21:55:38Z |
| description | The paper considers the task of crowd navigation monitoring, which might be performed using various sensors and technologies, with surveillance cameras being the most commonly employed. These cameras provide a video stream that typically lacks supplementary information. Extracting additional data from these streams could significantly enhance pedestrian behavior modeling and the automation of the monitoring process. A critical parameter in the analysis of pedestrian movement is their speed. The analytical method and the algorithm of pedestrians’ speed estimation based on the surveillance camera video are proposed. The first step of the proposed algorithm is object detection and tracking between frames. The second step is the speed estimation method, which is based on calculating the real-world distances and knowing camera parameters and distances in pixels on the resulting image. Implementation of the algorithm was tested on real videos and showed an error of about 0.04 m/s. |
| doi_str_mv | 10.20535/SRIT.2308-8893.2024.4.03 |
| first_indexed | 2025-07-17T10:28:28Z |
| format | Article |
| fulltext |
Publisher IASA at the Igor Sikorsky Kyiv Polytechnic Institute, 2024
43 ISSN 1681–6048 System Research & Information Technologies, 2024, № 4
UDC 004.932:621.397](045)
DOI: 10.20535/SRIT.2308-8893.2024.4.03
CROWD NAVIGATION MONITORING DURING EMERGENCIES
O.L. TYMOSHCHUK, M.O. TISHKOV, V.G. BONDARENKO
Abstract. The paper considers the task of crowd navigation monitoring, which
might be performed using various sensors and technologies, with surveillance cam-
eras being the most commonly employed. These cameras provide a video stream
that typically lacks supplementary information. Extracting additional data from these
streams could significantly enhance pedestrian behavior modeling and the automa-
tion of the monitoring process. A critical parameter in the analysis of pedestrian
movement is their speed. The analytical method and the algorithm of pedestrians’
speed estimation based on the surveillance camera video are proposed. The first step
of the proposed algorithm is object detection and tracking between frames. The sec-
ond step is the speed estimation method, which is based on calculating the real-
world distances and knowing camera parameters and distances in pixels on the re-
sulting image. Implementation of the algorithm was tested on real videos and
showed an error of about 0.04 m/s.
Keywords: computer vision, object tracking, object speed, video surveillance.
INTRODUCTION
Internal navigation monitoring in areas of large crowds during emergencies requires
advanced methods of computer vision, object tracking, and estimating flow speed.
Over the past decade, there have been numerous researches aimed at creating
intelligent video surveillance systems. Some typical applications are listed as follows:
public areas such as colleges, campuses, and governmental buildings;
traffic monitoring;
crowd management and analysis;
home security and intrusion detection;
home care and safety;
public transport areas such as airports, seaports, and bus/train terminals;
pedestrian detection and autonomous cars;
remote military surveillance, border monitoring, perimeter surveillance
for power plants, companies, etc [1].
One of the tasks of intelligent surveillance systems is collecting and process-
ing data about objects’ behavior. One of the core characteristics of the objects is
their moving speed. This paper focuses on estimating pedestrians’ speed, how-
ever, the same approach could be used for the estimation of any moving objects.
Considering pedestrian threads, this data can help develop more effective and safe
infrastructures during urban planning and transport engineering. In computer
modeling and human behavior simulation, moving speed is a critical parameter
for creating realistic agents, that reproduce human behavior in different use cases,
like evacuation or interaction with other pedestrians. Thereby, data about the
O.L. Tymoshchuk, M.O. Tishkov, V.G. Bondarenko
ISSN 1681–6048 System Research & Information Technologies, 2024, № 4 44
moving speed of pedestrians could facilitate improvements in various technolo-
gies that improve different aspects of our lives.
Problem statement
This paper aims to develop the analytical method and the algorithm for estimating
pedestrian speed using a video from a regular surveillance camera.
In order to achieve the goal, the next steps need to be done:
Analysis of existing methods and approaches of detection, tracking, and
speed estimation of moving objects, particularly pedestrians.
Development of the method for estimating real-world distance, having an
image from the camera.
Implementation of the algorithm based on the developed method that can
perform detection, tracking, storing of the pedestrians’ tracked data, and estimat-
ing their speed.
Testing of the implementation on the model problems. Algorithm quality
evaluation.
Related papers
The computer vision has been investigated by numerous scientists worldwide
[1–3]. There are research and implementations for object detection and classifica-
tion in images and videos [1]. The implementation proposed by authors in this
article is based on utilizing various sensors in video surveillance systems. Infrared
or thermal cameras aid in detecting people or other objects with contrasting tem-
peratures with the surrounding environment, providing the technical capability for
object detection [1].
Radars and lidars also provide technical ability for object detection and accu-
rate speed measurement. However, such sensors are expensive and rarely used.
For instance, the number of surveillance cameras in London is about 127.000 [4],
but the number of cameras with radar for speed estimation is only about 1000
speed cameras with radar [5]. Therefore, during further research of the methods and
their implementations, we will rely on the existing radarless surveillance systems.
An alternative approach utilizing object tracking [2] proposes the extraction
and tracking of objects in a video with a stationary background based on motion
analysis within frames and representation of images and objects as sets of struc-
tural elements. Evaluation of this approach, according to the provided experimen-
tal data, reveals that the method does not consider a potential merging and loss of
one of the objects, leading to data distortion.
There are numerous studies on object detection and tracking based on back-
ground separation with some specific improvements [6–9]. However, this ap-
proach has essential drawbacks as explained above.
The methods based on spatial filtering look promising for contour detection
[3]. The implemented solution enables image adjustment for object contour detec-
tion. One of the advantages of this approach is the ability to simplify data process-
ing having comparably low technical complexity.
Shvandt and Moroz consider a specific case of recognition and tracking of
laboratory animals, mice, and fish specifically. The authors provide numerous
methods of pattern recognition, including background subtraction and filtering
Crowd navigation monitoring during emergencies
Системні дослідження та інформаційні технології, 2024, № 4 45
methods. Also, trainable models are considered, which are used for face recogni-
tion, for instance. Such models are more generic, meaning they can work in a
wider range of situations, although, they are not ideal as well and can make mis-
takes or lose tracking objects. It is worth mentioning, that cases reviewed by the
authors, — laboratory animal recognition and tracking, are specific and facilitate
working conditions for algorithms like background separation since in laboratory
conditions background might be homogenous and also there is a possibility to
obtain its image without tracking objects [10].
Khan proposes a method of pedestrian detection based on foreground seg-
mentation and calculating the speed on pixels [11]. The author focuses on pedes-
trian detection on the crosswalks and aims to detect slow speed movements. This
approach has similar limitations to the previous author’s. Also, the calculation
speed in pixels on the image plane is the first step of estimating the real-world
speed, which is explained further in this article.
Zhao and Li present a pedestrian tracking method combining a Histogram of
Oriented Gradients detection and particle filter and a method for the detection of
abnormal crowd activity [12].
In this article, on the stage of people recognition and tracking a well-known
model YOLOv7 is used [13]. This model is rapidly developed. It is built on a
complex deep-learning model based on a convolutional neural network, which
consists of several convolutional and pulling blocks, that subtract characteristics
of the image on a different scale level. The determined characteristics are further
used to predict the position, size, and class of the object in the image.
Despite the breadth of previous research, pedestrian flow speed measure-
ment has received comparatively little attention, despite its significance. Under-
standing the flow speed of pedestrians can help to detect unusual events like con-
gestion or emergencies. Additionally, it is vital information for developing
pedestrian traffic models, which are instrumental in architectural design and
evacuation planning.
Teknomo et al. introduced a data collection framework for analyzing pedes-
trian flow in [14]. Their system autonomously identifies moving objects, tracks
them, and records their positions along with timestamps. By leveraging this col-
lected data on individual movements, the system could discern various character-
istics of pedestrian traffic flow, including individual speeds, flow rates, average
speeds, and directions. However, it's worth noting that their approach is con-
strained to video sequences captured from a top-view camera and does not ensure
accurate results for videos recorded with lower angles.
Tordeux et al. use an artificial neural network for predicting pedestrian speed
in different situations like corridors or bottlenecks [15]. The result of the imple-
mentation is prediction, but not the estimation of the physical movements which
is not the same and could be used in different use cases and for different goals.
There are studies aimed at estimating pedestrian flow speed using portable
devices like phones. Guo et al. propose a method of pedestrian speed estimation
based on the human pose detected by sensors of smartphones [16]. Huang et al.
use Wi-Fi sniffers to catch Wi-Fi probe requests from mobile devices to estimate
pedestrian flow speed and number of people in the crowd [17].
The approach by Lee et al. [18] is closely related to the problem discussed in
this research. They proposed a speed estimation algorithm based on calculating
conversion factors between the angles on the camera image and real-world motion
O.L. Tymoshchuk, M.O. Tishkov, V.G. Bondarenko
ISSN 1681–6048 System Research & Information Technologies, 2024, № 4 46
vectors. Also, they track the heads of the people for the algorithm. However all
people have different heights and the vertical distance from the camera to the
tracking object is different for each pedestrian, so they propose a way to predict
this distance based on statistical data and crowd density. It adds additional com-
plexity and might be a root for inaccuracy in calculations. Additionally, when cal-
culating conversion factors the authors assume that the rectangle on the camera
image projects into a rectangle on the pedestrian movement plane although this is
wrong since during central projection, a rectangle is projected into a trapezoid,
which changes the provided calculations.
In the considered resources different approaches for moving object detection
are proposed. One of them satisfies our needs so it will be used without any
changes. However, the main focus of our research is the speed estimation. Method
and implementation that cover all our needs and requirements have not been
found. The method suggested by Teknomo et al. [14] doesn’t work with cameras
in their real-life positions, Lee et al. [18] propose a similar approach to ours, but
simplifies projection which increases the error of calculations, Tordeux et al. [15]
try to predict the speed instead of estimation, and Guo et al. [16] propose an inter-
esting, but completely different approach using the radio signals detected from the
pedestrians’ phones.
The suggested analytical method allows pedestrians’ speed estimation from
the video of the camera located above the pedestrians, knowing the camera pa-
rameters.
PEDESTRIAN SPEED ESTIMATION METHOD
In this section, the proposed method for pedestrian speed estimation will be ex-
plained. The method is based on the distance estimation between two points on
a real-world plane based on the points on the image plane. Knowing the distance
and time between estimations the moving speed could be calculated.
Experiment model
Fig. 1 shows the pedestrian’s movement from point A to point B. To calculate the
pedestrian speed between these two positions we need to know the distance and
time elapsed. Time could be defined from the video. The real-world distance we
aim to estimate.
Fig. 1. Pedestrian movement
Crowd navigation monitoring during emergencies
Системні дослідження та інформаційні технології, 2024, № 4 47
Using Cartesian coordinates and knowing the resolution of the image we can
calculate the distance in pixels.
In the Fig. 2 rectangle
0000 DCBA is the full view of the
camera, BD
— the vector of the
movement and the diagonal of
the rectangle ABCD .
Fig. 3 shows how the
rectangle from Fig. 2 is projected
to the real-world plane from the
camera point located above the
image plane.
In the result we need to
estimate size of the vector ppDB
on the real-world plane which is
a projection of the vector of the pedestrian movement BD from the image plane.
Mathematical formalization
The first thing that is worth mentioning is that the rectangle on the image plane,
located at an angle relative to a real-world plane, is being projected from a point
above the image plane to a trapeze on a real-world plane — Fig. 4, as shown
in Fig. 1.
We aim to calculate the size of the
vector s — the diagonal of the trapeze.
For better understanding, we will
use simplified figures — Fig. 5.
According to a known formula,
denoting the bigger base AD as a , we
have for both cases of Fig. 5:
))( ( 22 ctghahd . (1)
Fig. 2. Pedestrian movement formalization
Y
A0 B0
C0
Fig. 3. Projection from image plane to a real-world plane
Fig. 4. Camera image and movement vector
projection to a real-world plane
O.L. Tymoshchuk, M.O. Tishkov, V.G. Bondarenko
ISSN 1681–6048 System Research & Information Technologies, 2024, № 4 48
In order to find the required diagonal we need to find the height of the tra-
peze — h , its larger base — a , and the angle near the smaller base from which
the diagonal extends — .
The height of the trapeze
Depending on the location of points on the image plane, the calculation of height
will vary. It depends on the vertical position of each point relative to the center of
the image plane. There are 3 possible cases (Fig. 6).
Here:
C — camera position;
H — height of the camera position;
ji PP , — points of the vector on the image plane;
M — middle of the image plane;
— angle of deviation of the camera from the vertical;
i — angle of deviation of the point iP from image plane vertical;
j — angle of deviation of the point jP from image plane vertical.
Segment AB would be the height of the projected trapeze on the real-world
plane. Having all described input data the height of the trapeze for cases a, b, and
c could be calculated by the next formulas:
))(tg)(tg( ijHh (2)
))tg()(tg( ijHh (3)
))(tg)(tg( ijHh (4)
f
P
f
MP
iyyi
i
Δ
arctg
||
arctg , (5)
Fig. 5: Trapeze diagonal
Fig. 6: Trapeze height projection
Crowd navigation monitoring during emergencies
Системні дослідження та інформаційні технології, 2024, № 4 49
where iPΔ — deviation of point iP from the center of the image plane along the
y-axis, f — focal length of the sensor.
Knowing the matrix dimensions, we can calculate the metric size of the pixel:
length
length
Matrix
Res
k . (6)
We can measure the deviation jiP ,Δ in pixels. By multiplying the vector’s
size in pixels by the size of the pixel k we can calculate the metric size of the
vector jiP ,Δ .
The larger base of the trapeze
A larger base of a trapeze will always be the projection of the upper side of the
rectangle on the image plane. In Fig. 4 11BA is
the top side of the rectangle on the image plane;
ABa — the larger side of the trapeze;
HCO — the height of the camera position,
COOB 11 and COOA 11 .
Since the triangles AOB and 111 BOA
on the Fig. 7 are similar:
h
kPPH
CO
BACO
a xjxi ||
1
11
(7)
Depending on the position of the point on
the image plane regarding the center of the
plane there are 2 possible cases: when the point is above or below the center of
the image — Fig. 8.
According to Fig. 8, h could be found using the following equation:
f
P
f
P
f
CPh
i
i
ii
arctgcos
arctgcos
)ΘΘ(cos (8)
C
Fig. 7. Larger base projection
ii
Fig. 8. Point position on the image plane
O.L. Tymoshchuk, M.O. Tishkov, V.G. Bondarenko
ISSN 1681–6048 System Research & Information Technologies, 2024, № 4 50
Equations (7) and (8) give us the total equation for calculating the larger
base of the trapeze:
f
P
f
f
P
kPPH
a
i
i
xjxi
arctgcos
arctgcos||
, (9)
— the angle near the small base of the trapeze
Fig. 9 shows the same projection as Fig. 3 but with required additions.
In this section, we will consider triangle
100 OBA pp from Fig. 9 in more detail — Fig. 10.
QS — projection of some vector on the
image plane. We aim to find — the angle
near a small base of the trapeze.
NP
PO
NPO 1
1 arctg , (10)
NP could be calculated using (9), PO1 con-
sists of three parts
11 OOMOPMPO , (11)
PM could be calculated using (2)
)Θ(tgHMO , (12)
)Θ(ctgΘ
2
tg1 HHOO
. (13)
Considering equations above:
)Θ(cos
)Θ(Θcos)(ΘΘctg)Θ(Θtg(
arctg maxmax
kPNH
H
xx
, (14)
where maxΘ — maximum vertical deflection angle or half of the vertical viewing
angle of the camera.
All required components for (1) could be found using the equations provided
above. The required input parameters are:
Fig. 9. Extended projection
Fig. 10. Projection plane
Crowd navigation monitoring during emergencies
Системні дослідження та інформаційні технології, 2024, № 4 51
H — the height of the camera position; f — the focal length of the cam-
era; Θ — the tilt angle of the camera; — the vertical viewing angle of the
camera.
ALGORITHM IMPLEMENTATION
Detection and tracking
The first step of the implementation of pedestrians’ speed estimation is pedestri-
ans’ identification and tracking. The YOLOv7 object detection model variant by
Rizwan Munawar [19] is used in the proposed implementation.
Pedestrians are detected on each frame of the video and data is saved to the
MySQL database for further processing. After the video file is processed, the
speed estimation step can be executed. It updates the database with speed data,
calculated using the provided algorithm, foreach sN ' frame. As a result, we have
data that could be displayed on the video or used for further research, for instance,
for estimating the pedestrians’ average speed.
The algorithm requires two input parameters, H and Θ , that can be meas-
ured manually and the measurement error can significantly impact the perform-
ance. To improve the accuracy, a calibration step is implemented that can be exe-
cuted between detection and speed estimation steps. It requires an array of vectors
on the image plane and the distance of their projections on the real-world plane.
On the calibration step parameters are tuned to decrease the root-mean-square
deviation of calculated results of provided vectors and their real distance.
Final algorithm
At the estimation step, we have all the required data:
Coordinates of pedestrians at every frame;
Camera parameters:
– height of the camera position — h;
– the tilt angle of the camera — Θ ;
– the focal length of the camera in millimeters;
– camera resolution;
– sensor size
– vertical viewing angle of the camera — .
The estimation algorithm calculates the speed between every N frames,
where N could be set arbitrarily. Knowing the FPS of the video the time between
frames could be calculated and the distance traveled by the pedestrian in N frames
could be calculated using equation (1), substituting parameters from equations
(2–4), (7), (14). Dividing the founded distance by time between N frames gives us
speed. Operation is repeated for every N frames till the end of the video for each
tracked pedestrian on the video and the results are stored in the database.
Results and model problems testing
Testing the algorithm performance on real videos is a problem since the actual
speed of pedestrians is not known. Testing videos were recorded with known dis-
tances and times to evaluate the algorithm’s performance. The results of the esti-
mation and real manually measured values are presented in Table.
O.L. Tymoshchuk, M.O. Tishkov, V.G. Bondarenko
ISSN 1681–6048 System Research & Information Technologies, 2024, № 4 52
Test video results
Test # Real average speed Estimated average speed Error
1 1.17 m/s 1.25 m/s 6.8%
2 1.1 m/s 1.09 m/s 1%
3 1.19 m/s 1.22 m/s 2.5%
4a 0.99 m/s 0.99 m/s 0%
4b 0.7 m/s 0.77 m/s 11%
In Fig. 11 the estimation process is shown. As we can see, due to the chang-
ing size of the detected object, the tracked trajectory for a certain perspective ac-
quires a stepped form. It might increase the error, however, as we can see from
the results, this error is compensated through the entire path.
As mentioned above, the proposed algorithm requires a set of parameters of
the processed video to be set. Without them, it will work but the results might dif-
fer from real speeds significantly. An example of such a case is shown on a video
from the Internet where the parameters of the camera are unknown, which is
shown in Fig. 12.
Also, for videos with known parameters, we can expect realistic estimations
only for pedestrians moving on one plane, which is used for measuring the height
Fig. 11: Estimation process
Fig. 12. Speed detection with unknown parameters
Crowd navigation monitoring during emergencies
Системні дослідження та інформаційні технології, 2024, № 4 53
of the camera position. The results for other detected pedestrians will be wrong.
The detected pedestrian on the balcony at the left top corner of Fig. 12 can be
used as an example of such a case. He is standing above the plain where other
people walking, so even having the correct parameters of the camera won’t allow
us to estimate his speed.
CONCLUSION
In this paper, the method for pedestrian speed estimation based on the video from
a surveillance camera, knowing camera parameters is proposed. The proposed
method estimates distance on the real-world plane by visible movement vectors
on the image plane. In the proposed implementation, the existing model for pe-
destrian detection and tracking is used. Unlike the previous methods, the proposed
method doesn’t have an issue with different heights of the tracked objects, since it
uses points on the surface for tracking and estimation.
To evaluate the algorithm’s accuracy, it was run on testing videos with
known pedestrian speed. The resulting relative error is 0–11%, while in absolute
terms, it ranges from 0 to 0.08 m/s.
Next steps. Knowing the parameters of the surveillance camera and having
access to its video proposed algorithm could be used to retrieve data for pedestri-
ans’ behavior analysis. This information could be used during crowd modeling in
decision support systems. Also, by improving performance or running on more
powerful hardware proposed algorithm may work in a real time. In this case, it
could be used for detecting unusual behavior in the crowd.
The current implementation can be found on GitHub [20].
REFERENCES
1. S.W. Ibrahim, “A comprehensive review on intelligent surveillance systems,” CST, vol.
1, no. 1, pp 7–14, May 2016.
2. O.V. Aharkov, “Object detection and tracking by motion analysis,” Shtuchnyi intelekt,
no. 1-2, pp. 28–35, 2015.
3. O.V. Bahatskyi, “The instrumental complex for research algorithms for object bounda-
ries,” Computers means, networks and systems, no. 15, pp. 74–78, 2016.
4. P. Bischoff, Surveillance camera statistics: which cities have the most CCTV cameras?
Accessed on: May 23, 2023. [Online]. Available: https://www.comparitech. com/vpn-
privacy/the-worlds-most-surveilled-cities/
5. A. Moody, “Speed Camera Hotspots,” GoShorty. Accessed on: Jan 17, 2022. [Online].
Available: https://goshorty.co.uk/blog/speed-camera-hotspots/
6. R.D. Sharma, S.K. Gupta, “A Survey on Moving Object Detection and Tracking Based
On Background Subtraction,” Oxford Journal of Intelligent Decision and Data Science, vol.
2018, pp. 55–62, 2018. doi: https://doi.org/10.5899/2018/ojids-00041
7. P. Spagnolo, T. Orazio, M. Leo, and A. Distante, “Moving object segmentation by back-
ground subtraction and temporal analysis,” Image and Vision Computing, vol. 24, no. 5,
pp. 411–423, 2006. doi: https://doi.org/10.1016/j.imavis.2006.01.001
8. A. Hardas, B.M. Dattatray, and M. Vibha, “Moving Object Detection using Background
Subtraction Shadow Removal and Post Processing,” International Journal of Computer
Applications, ICCT 2015, pp. 1–5, 2015.
9. P. Kalaivani, S. Vimala, “Human action recognition using background subtraction
method,” International Research Journal of Engineering and Technology, vol. 2,
pp. 1032–1035, 2015.
10. M.A. Shvandt, V.V. Moroz, “Overview of the detection and tracking methods of the lab
animals,” System Research and Information Technologies, no. 1, pp. 124–148, Apr 2022.
doi: https://doi.org/10.20535/SRIT.2308-8893.2022.1.10
O.L. Tymoshchuk, M.O. Tishkov, V.G. Bondarenko
ISSN 1681–6048 System Research & Information Technologies, 2024, № 4 54
11. D. Khan, “Estimating Speeds and Directions of Pedestrians in Real-Time Videos:
A solution to Road-Safety Problem” in AgeingAI 2013 The Challenge of Ageing Society:
Technological Roles and Opportunities for Artificial Intelligence, Turin, Italy, 2013.
12. F. Zhao, J. Li, “Pedestrian Motion Tracking and Crowd Abnormal Behavior Detec-
tion Based on Intelligent Video Surveillance,” Journal of Networks, no. 10, pp. 2598–
2605, 2014. doi: https://doi.org/10.4304/jnw.9.10.2598-2605
13. B.Y. Wang, A. Bochkovskiy, and H.Y.M. Liao, “YOLOv7: Trainable bag-of-freebies
sets new state-of-the-art for real-time object detectors,” in CVPR, New Orleans, LA, USA,
2022, pp. 7464–7475. doi: https://doi.org/10.48550/arXiv.2207.02696
14. K. Teknomo, “Microscopic Pedestrian Flow Characteristics: Development of an Image
Processing Data Collection and Simulation Model,” Ph.D. dissertation, Department of
Human Social Information Sciences, Graduate School of Information Sciences, Tohoku
University, Japan, 2002.
15. A. Tordeux, M. Chraibi, A. Seyfried, and A. Schadschneider, “Prediction of Pedestrian
Speed with Artificial Neural Networks,” in Proc. Int. Conf. Traffic Granular Flow, pp.
327–335, 2017. doi: https://doi.org/10.1007/978-3-030-11440-4_36
16. G. Guo et al., “A Pose Awareness Solution for Estimating Pedestrian Walking Speed,” Re-
mote Sensing, vol. 11, no. 1, pp. 55–73, 2019. doi: https://doi.org/10.3390/rs11010055
17. B. Huang, G. Mao, Y. Qin, and Y. Wei, “Pedestrian flow estimation through passive
WiFi sensing,” IEEE Trans. on Mobile Computing, vol. 20, no. 4, pp. 1529–1542, 2021.
doi: https://doi.org/10.1109/TMC. 2019.2959610
18. G.G. Lee, K.H. Ka, and W.Y. Kim, “Estimation of pedestrian flow speed in surveillance
videos,” in Conf. Korean Society of Broadcast Engineers Conference, Seoul, Republic of
Korea, 2009, pp. 330–333.
19. R. Munawar, YOLOv7 object tracking. 2023. [Online]. Available: https://github.com/
RizwanMunawar/yolov7-object-tracking
20. M. Tishkov, Pedestrian speed estimation. 2024. [Online]. Available: https://github.com/
MaksymTishkov/pedestrian_speed_estimation
Received 26.04.2024
INFORMATION ON THE ARTICLE
Oksana L. Tymoshchuk, ORCID: 0000-0003-1863-3095, Educational and Research
Institute for Applied System Analysis of the National Technical University of Ukraine
“Igor Sikorsky Kyiv Polytechnic Institute”, Ukraine, e-mail: oxana.tim@gmail.com
Maksym O. Tishkov, Educational and Research Institute for Applied System Analysis of
the National Technical University of Ukraine “Igor Sikorsky Kyiv Polytechnic Institute”,
Ukraine, e-mail: maksym.tishkov@gmail.com
Victor G. Bondarenko, ORCID: 0000-0003-1663-4799, Educational and Research Insti-
tute for Applied System Analysis of the National Technical University of Ukraine “Igor
Sikorsky Kyiv Polytechnic Institute”, Ukraine, e-mail: bondarenvg@gmail.com
МОНІТОРИНГ НАВІГАЦІЇ НАТОВПУ ПІД ЧАС НАДЗВИЧАЙНИХ
СИТУАЦІЙ / О.Л. Тимощук, М.О. Тішков, В.Г. Бондаренко
Анотація. Розглянуто задачу моніторингу навігації натовпу, що може здійс-
нюватися за допомогою різних сенсорів і технологій, причому найчастіше ви-
користовуються камери спостереження. Ці камери забезпечують відеопотік,
який зазвичай не містить додаткової інформації. Отримання додаткових даних
з цих відеопотоків може значно покращити моделювання поведінки пішоходів
та автоматизацію процесу моніторингу. Критичним параметром в аналізі руху
пішоходів є їх швидкість. Запропоновано аналітичний метод та алгоритм оці-
нювання швидкості пішоходів на основі відео з камер спостереження. Першим
кроком запропонованого алгоритму є розпізнавання об’єктів та їхній трекінг
між ферймами відео. Наступний крок — оцінювання швидкості руху об’єктів,
що базується на розрахунку реальних відстаней з відомими параметрами каме-
ри та відстанями в пікселах на результуючому зображенні. Додатково пропо-
нується алгоритм калібрування для вирівнювання параметрів камери з метою
забезпечення найточніших результатів. Реалізацію алгоритму протестовано на
реальних відео, похибка — близько 0,04 м/с.
Ключові слова: комп’ютерний зір, трекінг об’єктів, швидкість руху об’єктів,
відеоспостереження.
|
| id | journaliasakpiua-article-302774 |
| institution | System research and information technologies |
| keywords_txt_mv | keywords |
| language | English |
| last_indexed | 2025-07-17T10:28:28Z |
| publishDate | 2024 |
| publisher | The National Technical University of Ukraine "Igor Sikorsky Kyiv Polytechnic Institute" |
| record_format | ojs |
| resource_txt_mv | journaliasakpiua/4e/c189e8c665b382bdaf08453b4cc7924e.pdf |
| spelling | journaliasakpiua-article-3027742025-02-09T21:55:38Z Crowd navigation monitoring during emergencies Моніторинг навігації натовпу під час надзвичайних ситуацій Tymoshchuk, Oksana Tishkov, Maksym Bondarenko, Victor комп’ютерний зір трекінг об’єктів швидкість руху об’єктів відеоспостереження computer vision object tracking object speed video surveillance The paper considers the task of crowd navigation monitoring, which might be performed using various sensors and technologies, with surveillance cameras being the most commonly employed. These cameras provide a video stream that typically lacks supplementary information. Extracting additional data from these streams could significantly enhance pedestrian behavior modeling and the automation of the monitoring process. A critical parameter in the analysis of pedestrian movement is their speed. The analytical method and the algorithm of pedestrians’ speed estimation based on the surveillance camera video are proposed. The first step of the proposed algorithm is object detection and tracking between frames. The second step is the speed estimation method, which is based on calculating the real-world distances and knowing camera parameters and distances in pixels on the resulting image. Implementation of the algorithm was tested on real videos and showed an error of about 0.04 m/s. Розглянуто задачу моніторингу навігації натовпу, що може здійс-нюватися за допомогою різних сенсорів і технологій, причому найчастіше ви-користовуються камери спостереження. Ці камери забезпечують відеопотік, який зазвичай не містить додаткової інформації. Отримання додаткових даних з цих відеопотоків може значно покращити моделювання поведінки пішоходів та автоматизацію процесу моніторингу. Критичним параметром в аналізі руху пішоходів є їх швидкість. Запропоновано аналітичний метод та алгоритм оцінювання швидкості пішоходів на основі відео з камер спостереження. Першим кроком запропонованого алгоритму є розпізнавання об’єктів та їхній трекінг між ферймами відео. Наступний крок — оцінювання швидкості руху об’єктів, що базується на розрахунку реальних відстаней з відомими параметрами камери та відстанями в пікселах на результуючому зображенні. Додатково пропонується алгоритм калібрування для вирівнювання параметрів камери з метою забезпечення найточніших результатів. Реалізацію алгоритму протестовано на реальних відео, похибка — близько 0,04 м/с. The National Technical University of Ukraine "Igor Sikorsky Kyiv Polytechnic Institute" 2024-12-25 Article Article application/pdf https://journal.iasa.kpi.ua/article/view/302774 10.20535/SRIT.2308-8893.2024.4.03 System research and information technologies; No. 4 (2024); 43-54 Системные исследования и информационные технологии; № 4 (2024); 43-54 Системні дослідження та інформаційні технології; № 4 (2024); 43-54 2308-8893 1681-6048 en https://journal.iasa.kpi.ua/article/view/302774/312890 |
| spellingShingle | комп’ютерний зір трекінг об’єктів швидкість руху об’єктів відеоспостереження Tymoshchuk, Oksana Tishkov, Maksym Bondarenko, Victor Моніторинг навігації натовпу під час надзвичайних ситуацій |
| title | Моніторинг навігації натовпу під час надзвичайних ситуацій |
| title_alt | Crowd navigation monitoring during emergencies |
| title_full | Моніторинг навігації натовпу під час надзвичайних ситуацій |
| title_fullStr | Моніторинг навігації натовпу під час надзвичайних ситуацій |
| title_full_unstemmed | Моніторинг навігації натовпу під час надзвичайних ситуацій |
| title_short | Моніторинг навігації натовпу під час надзвичайних ситуацій |
| title_sort | моніторинг навігації натовпу під час надзвичайних ситуацій |
| topic | комп’ютерний зір трекінг об’єктів швидкість руху об’єктів відеоспостереження |
| topic_facet | комп’ютерний зір трекінг об’єктів швидкість руху об’єктів відеоспостереження computer vision object tracking object speed video surveillance |
| url | https://journal.iasa.kpi.ua/article/view/302774 |
| work_keys_str_mv | AT tymoshchukoksana crowdnavigationmonitoringduringemergencies AT tishkovmaksym crowdnavigationmonitoringduringemergencies AT bondarenkovictor crowdnavigationmonitoringduringemergencies AT tymoshchukoksana monítoringnavígacíínatovpupídčasnadzvičajnihsituacíj AT tishkovmaksym monítoringnavígacíínatovpupídčasnadzvičajnihsituacíj AT bondarenkovictor monítoringnavígacíínatovpupídčasnadzvičajnihsituacíj |