Volume 9, Issue 6, June – 2024 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24JUN071

A Proposed Model for Fainting People Detection

Using Media Pipe Technology
1
Areege Samir Elhosany; 2Salma Ahmed Khalil; 3Mohamed Saied El-Sayed Amer
4
Sayed Mohamed; 5Ahmed Atef Abdelhamid; 6Abdelrahman Samir; 7Mohamed Mansour
3
Assitant Professor, 1,2,3,4,5,6,7Computer Science Canadian International College New Cairo, Egypt

Abstract:- Nowadays, falls put patients' safety and health The suggested model integrates a camera system with
in danger in hospitals and assisted living facilities, Google's Media Pipe technology[7] and a mobile application
especially at night. The camera system used in this study interface that is easy to use in order to improve patient safety
is the suggested model; it records video footage and, in in healthcare institutions.
real time, analyzes patterns using Media Pipe algorithms
to detect fainting or falls. An instant alert is generated The suggested model operates by analyzing the frame
and sent to a mobile app connected to the camera system as the camera system records the senior citizens' movements
if such an event is detected. The caretakers, nurses, or in order to identify the fainting event as soon as it happens.
security officers entrusted with providing assistance can Next, begin alerting the appropriate parties so they can take
communicate with each other using the smartphone the required steps to assist the individual who is fainting.
application. They follow the message to a specific
location. Lastly, the system aims to improve the quality The suggested solution comes with a smartphone
of care and support for the senior population by application that makes it easier for the camera system and
accelerating the response to instances of fainting and the accountable parties to communicate via an internal
falling through the use of computer vision technologies network.
and real-time notifications.
This work's remaining portions are arranged as follows.
Keywords:- Falling People, Human Detection, Media Pipe The related work is discussed in Section 2. The methodology
Fainting People, Elder Care. and suggested algorithm for the faint detection model are
explained in Section 3. The experimental findings and the
I. INTRODUCTION comments are included in Section 4. Section 5 concludes the
study and discusses future research.
Elderly people, patients, and pregnant women are
prone to fainting, which can lead to both physical and II. RELATED WORK
emotional health problems. When someone falls and hits
their head on the floor or another hard surface, they usually The breadth and applications of these human detection
faint. Reaction and rescue time are critical components of and position-related initiatives can vary greatly, and they
emergency medical therapy for fainting patients. Therefore, frequently make use of a variety of technologies, including
it is essential to identify these cases as soon as possible in sensor-based systems, computer vision, and machine
order to treat this population quickly. learning. Each of these domains embodies unique
technological applications, utilizing methods and
A number of methods have been put out to identify fall instruments such as computer vision, deep learning, and
motion. Using a press button on a wearable is one such sensor-based systems to tackle diverse issues related to
technique. This enables the person who has fallen to press human identification, tracking, and engagement in a range of
the button and request help. This strategy, however, is sectors and domains.
useless if the victim passes out right away. Additionally, an
alarm is automatically transmitted by a wearable feeble Target identification and tracking are important issues
motion sensor that detects motion using the tilt and for wireless sensor networks and are vital to public safety.
acceleration sensors [1]. For the elderly, patients, and even expectant mothers, faint
detection is a critical concern. It is widely used in
Around the world, human pose estimation technology contemporary society. A technique for obtaining data on the
is being actively researched in a variety of fields, including location and behavior of faint events in the sensing
sports, entertainment, home senior care, gesture control, environment was made possible by this work. This method
surveillance, job monitoring, and even metaverse avatars. uses the Kalman filter in conjunction with the Camshift
Human pose estimation techniques are generally divided into tracking algorithm to locate and follow faint persons. Trials
two categories: 2D and 3D coordinate estimation showed that the method performs well in difficult
techniques[2-6]. circumstances for both tracking and detection[8].

IJISRT24JUN071 www.ijisrt.com 803

Volume 9, Issue 6, June – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24JUN071

For usage in a thermal imaging camera to track fainting proposed technique improves the unified detecting CNN of
incidents, a different straightforward faint detection the YOLO model. The speed-up benefit obtained from using
technique was presented. Moreover, this method makes it 2-D input data consisting of grey-level image data and
possible to start therapy right away in order to save a person's foreground context information instead of 3-D color input
life when they faint. The suggested surveillance system can data offsets the increased computing time caused by
identify human faint events with a high accuracy of 96.15% additional AGMM processing [10].
in dimly lit areas and 86.19% inside, according to
experimental results on sample images[9]. III. METHODOLOGY

In a different study, they present a novel real-time The open-source, cross-platform MediaPipe Pose
human detection method based on CNN with unified (MPP) framework from Google is used in this study to
detection and AGMM for visual surveillance employing estimate the 2D human joint coordinates in each frame of an
omni-directional cameras. contrasted with state-of-the-art image. MediaPipe Pose uses machine learning (ML) to
CNN-based object detection techniques. Although it is not design pipelines and manage cognitive data based on videos.
as accurate, the YOLO model-based object identification Figure 1 illustrates how MPP extracts 2D landmarks from
method is quite quick. With additional foreground the human body using a BlazePose[11].
contextual information obtained from pre-stage AGMM, the

Fig 1: Definition of Landmarks in MediaPipe Pose[12]

BlazePose is a lightweight machine learning

architecture that uses CPU inference to deliver real-time
performance on PCs and mobile devices. To estimate a pose
using normalized coordinates, multiply the inverse ratio by
the values of the y-axis pixels. Utilizing indices 11, 12, 13,
14, 15, 16, 23, 24, 25, 26, 27, and 28, we calculated arbitrary
postures and motions utilizing 12 of the estimated MPP
landmarks, as seen in the Figure 1 below.

As seen in Figure 2, prompt alerts are generated and

sent to healthcare professionals via alarms, emails, text
messages, or notifications on an intuitive mobile application
interface when a possible fall event is identified, enabling
them to take the appropriate action. Fig 2: Proposed Model Architecture

As seen in Figure 2, prompt alerts are generated and

sent to healthcare professionals via alarms, emails, text
messages, or notifications on an intuitive mobile application
interface when a possible fall event is identified, enabling
them to take the appropriate action.

IJISRT24JUN071 www.ijisrt.com 804

Volume 9, Issue 6, June – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24JUN071

IV. EXPERIMENTS AND RESULTS As seen in two different images, our proposed model's
correction procedure involves adjusting important points
Python is used to build the experiment and test of the along straight lines in the image, starting with an initial
proposed model, and a high-resolution camera is used to estimated pose (left) and moving forward to the ground truth
accurately detect the human position. position y (right). The effectiveness of activity estimation
was evaluated using a sudden fall motion estimation; Figure
3 shows the 2D posture estimation.

Fig 3: Results of 2D pose estimation obtained by MPP for a sequence of images in a Sudden[7]

And using MediaPipe Pose and camera images, the 3D parameters related to stable stances, makes the reconstructed
stances were generated from the 2D stances. Figure 4 humanoid models resemble the real 3D postures.
illustrates how the loss function, which reflects a number of

Fig 4: Results of the Reconstructed 3D Humanoid Poses [7]

IJISRT24JUN071 www.ijisrt.com 805

Volume 9, Issue 6, June – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24JUN071

Because each key point requires a consistent set of [7]. Kim, Jong-Wook & Choi, Jin-Young & Ha, Eun-Ju &
corrections across images—the final correction being a scaled Choi, Jae-Ho. (2023). Human Pose Estimation Using
version—this method makes testing predictions easier. MediaPipe Pose and Optimization Method Based on a
Similar to a control system, the model's feedback Humanoid Model. Applied Sciences. 13. 2700.
mechanisms allow it to detect when the solution is getting 10.3390/app13042700.
close to precision and minimize excessive movement of [8]. Wang, Zhen-hai & Xu, Bo. (2015). A Robust Home
crucial points. To improve clarity and simplify the visuals, Alone Faint Detection Based on Wireless Sensor
only a selection of the most important locations are shown Networks. International Journal of Distributed Sensor
with linear trajectories. Networks. 2015. 1-5. 10.1155/2015/534980.
[9]. W. K. Wong, H. L. Lim, C. K. Loo and W. S. Lim,
A smartphone application included with the suggested "Home Alone Faint Detection Surveillance System
system makes it easier for the monitoring camera and the Using Thermal Camera," 2010 Second International
rescue teams to communicate and take appropriate action in Conference on Computer Research and Development,
the event that a fainting occurrence is identified. Kuala Lumpur, Malaysia, 2010, pp. 747-751, doi:
10.1109/ICCRD.2010.163.
V. DISCUSSION AND FUTURE WORK [10]. Nguyen, T.B., Nguyen, V.T., Chung, S., & Cho, S.
(2016). Real-time Human Detection under Omni-dir
This research presents a fainting detection model that ectional Camera based on CNN with Unified
uses Media Pipe technology to identify fainting individuals. Detection and AGMM for Visual Surveillance.
The model also includes a mobile application as a component Journal of Korea Multimedia Society, 19, 1345-1360.
that helps with the falling event rescue process. [11]. Bazarevsky, V.; Grishchenko, I. On-Device, Real-
Time Body Pose Tracking with MediaPipe BlazePose,
The model makes advantage of Google Technology's Google Research. Available online:
media pipe technology, which provides a good estimate of the https://ai.googleblog.com/2020/08/on-device-real-
human stance. time-body-pose-tracking.html (accessed on 10 August
2021).
As a result, our suggested system will alert the relevant [12]. MediaPipe Pose. Available online:
units to take appropriate action if many humans are identified https://google.github.io/mediapipe/solutions/pose.ht
on such an application site. However, the approach proposed ml (accessed on 28 December 2021).
in this study needs to be refined in order to be applicable in
scenarios when a person who has fainted is ignored by those
around them, such as in a homicide investigation. One way to
find out if someone is conscious or unconscious is to follow
them in a single picture. These topics will be explored in
further work.

REFERENCES

[1]. So, I., Han, D., Kang, S., Kim, Y., & Jung, S. (2008).
Recognition of Fainting Motion from Fish-eye Lens
Camera Images.
[2]. Sarafianos, S.; Boteanu, B.; Ionescu, B.; Kakadiaris,
I.A. 3D human pose estimation: A review of the
literature and analysis of covariates. Comput. Vis.
Image Underst. 2016,152, 1–20. [CrossRef]
[3]. Chen, Y.; Tian, Y.; He, M. Monocular human pose
estimation: A survey of deep learning-based methods.
Comput. Vis. Image Underst. 2020,192, 102897.
[CrossRef]
[4]. Wang, J.; Tan, S.; Zhen, X.; Xu, S.; Zheng, F.; He, Z.;
Shao, L. Deep 3D human pose estimation: A review.
Comput. Vis. Image Underst. 2021,210, 103225.
[CrossRef]
[5]. Yurtsever, M.M.E.; Eken, S. BabyPose: Real-time
decoding of baby’s non-verbal communication using
2D video-based pose estimation. IEEE Sens. 2022,22,
13776–13784. [CrossRef]
[6]. Alam, E.; Sufian, A.; Dutta, P.; Leo, M. Vision-based
human fall detection systems using deep learning: A
review. Comput. Biol. Med. 2022,146, 105626.
[CrossRef]

IJISRT24JUN071 www.ijisrt.com 806