A

REPORT
ON

SLEEP POSTURE RECOGNITION SYSTEM

Name:- id:-
HARIKARAN KUMARAVEL 2021B4A33180H
TRISHA S IYER 2022A8PS2028H
MOHANA KEERTHI 2022A8PS0738H

BIRLA INSTITUTE OF TECHNOLOGY AND SCIENCE

HYDERABAD

NOVEMBER 2024
 Index

INTRODUCTION 3

ABSTRACT 4

PRIOR WORK 5

PROBLEM STATEMENT 6

EVALUATION OF WORK 7

FINAL SOLUTION 8-12

HARDWARE 12-14

CONCLUSION 15
 INTRODUCTION

Sleep posture recognition plays a critical role in several

fields, including healthcare, particularly in diagnosing
conditions like sleep apnea, preventing pressure ulcers in
bedridden patients, and ensuring safe postures for epilepsy
patients. While traditional methods like wearable devices
have been used, they present limitations. Wearables may
cause discomfort which makes it impossible to get accurate
data as the test subject’s sleep is disturbed and hence
inauthentic .
A systematic review of each of our research papers
revealed a few recurring challenges. For each challenge, a
potential solution was formulated. These solutions were
then integrated into a cohesive, unified solution.
 ABSTRACT

Our project aims to develop a novel sleep posture

recognition system that overcomes the limitations of existing
methods. By combining a pressure-sensitive bed and a
camera-based system, we seek to provide a more accurate
and comprehensive assessment of sleep postures. The
pressure-sensitive bed offers precise information about the
distribution of pressure across the sleep surface, enabling the
identification of gross postural changes. Simultaneously, the
camera-based system, optimized for low-light conditions,
captures subtle postural details that may be missed by
pressure sensors alone.
By integrating these two modalities and employing
advanced machine learning techniques, our system can
accurately classify a wide range of sleep postures,
including those that are challenging for traditional
methods. This enhanced accuracy allows for a more
precise analysis of sleep quality and the identification of
potential sleep disorders. Furthermore, our system is
designed to be user-friendly and non-invasive, minimizing
discomfort and privacy concerns.
Prior work in the papers we reviewed:-

Wearables: Devices like smartwatches disrupt sleep due to

discomfort.

Computer Vision: Effective, but privacy concerns and

occlusions (like blankets) remain issues.

RF-based Systems: While non-intrusive, these systems

struggle with accuracy and deployment.

Radar-based Approaches: Early radar systems were

promising but had trouble recognizing symmetrical
postures and required long training times.

Peripheral Environment Sensors: These systems use

sensors around the environment to track human motion, but
they often lack accuracy and complexity.

Polysomnography (PSG): The gold standard in clinical

sleep assessments,PSG captures multiple physiological
signals but is expensive, intrusive,and requires subjects to
wear devices, affecting natural sleep conditions
 PROBLEM STATEMENT
The primary limitations identified in the reviewed literature
were the following:
1. Limited Posture Recognition: The majority of
existing posture recognition systems are capable of
identifying only a limited number of sleep postures,
typically around six.
2. Lack of Directional and Supine/Prone
Differentiation: Many systems fail to distinguish
between left and right orientations and cannot
accurately determine whether a person is sleeping
face-up or face-down.
3. Limited Operational Capability in Low-Light
Conditions: Existing camera-based systems are often
hindered by the low-light conditions typical of sleep
environments, leading to reduced accuracy and
reliability.
4. Insufficient Data for Pediatric Applications: The
scarcity of data on infant sleep postures, largely due to
ethical and practical considerations, hinders the
development and validation of accurate posture
recognition systems for this vulnerable population.
This however is a problem that is beyond the scope of
this presentation as obtaining pediatric data is not in
our realm of possibility.
Evaluation of Solutions with Prior Solutions:-

Non-Intrusiveness
The pressure-sensor-based mattress allows subjects to sleep
naturally without wearing any devices or being monitored
by cameras or microphones. This reduces discomfort and
privacy concerns.Polysomnography (PSG) requires
multiple physiological sensors attached to the subject's
body,causing discomfort. Wearable devices like wristbands
and patches still involve contact, which can interfere
with natural sleep.
Accuracy
Achieves high accuracy in posture detection, with 99.986%
accuracy for sitting vs. lying postures and
96.987% for identifying sleep postures (face-up,
face-down, side positions). This level of precision
outperforms many prior approaches using pressure-sensing
technology.Wearable devices like wristbands
track sleep but are less reliable in determining exact
postures.
Cost and Convenience
The smart mattress system is more affordable and suitable
for home use, enabling long-term monitoring in
a natural environment. PSG is expensive and confined to
clinical settings.
FINAL SOLUTION
To overcome the limitations of detecting only six
predefined sleep positions, we propose the installation of an
11 x 20 pressure mat beneath the mattress, integrated into
the bed frame. This pressure mat captures detailed data on
the specific areas of the body in contact with the surface.
By collecting this information, the system can identify the
exact sleep positions with precision. The data is transmitted
to a connected computer for visualization and analysis,
providing a comprehensive understanding of body coverage
and eliminating the constraints of predefined position
detection.

Despite this improvement, a significant challenge remains

in differentiating between left and right orientations and
distinguishing face-up from face-down sleeping. To address
this, we propose the inclusion of an infrared (IR) camera
strategically positioned at either the foot or the head of the
bed, based on practical considerations such as convenience
and ease of setup. The IR camera is specifically tasked with
determining the direction in which the sleeper's head is
turned, providing crucial orientation data. It records and
stores only the head’s directional information, ensuring a
streamlined and efficient operation without unnecessary
data processing.
The IR camera also resolves the issue of operation in
low-light or dark environments, as it is capable of capturing
images regardless of ambient lighting conditions. Its ability
to function seamlessly in darkness makes it a reliable
component of the system.

By combining the advanced pressure mat with the targeted

capabilities of the IR camera, this dual-system approach
offers a comprehensive, accurate, and efficient solution for
detecting a wide range of sleep positions and orientations.
It significantly enhances the functionality of the sleep
tracking system, providing valuable insights for both
personal and clinical applications.

This figure shows the set-up of the camera on the bed.

Visual representation of the output of the pressure sensing
mat.
Integration of Camera and Machine Learning for
Enhanced Sleep Analysis :

To complement the pressure mat and address orientation

challenges, we propose the installation of an infrared (IR)
camera as a key component of the sleep analysis system.
This camera is designed to capture images at regular
intervals throughout the night, providing valuable visual
data for orientation determination. The camera feed is
integrated with a custom-designed code, enabling seamless
processing and analysis of both audio and video inputs.
The machine learning (ML) model embedded in the system
operates by capturing an image every 20 seconds. These
images are processed to identify and classify the sleeper’s
orientation, including whether they are lying on their right
or left side, face-up, or face-down. This capability provides
a clear and detailed understanding of the sleeper's
movements and postures throughout the night.

In addition to real-time processing, the system is designed

to store the captured data in an Excel spreadsheet for
further analysis. This includes storing both the orientation
classifications from the image analysis and audio data
collected during the night. The structured data enables
efficient review and facilitates detailed insights into the
sleeper's behavior over time.

This combined approach of visual and audio data analysis,

supported by an ML-driven framework, ensures a robust
and comprehensive sleep monitoring solution. By
leveraging automation and advanced analytics, this system
provides accurate, reliable, and actionable insights that can
significantly enhance personal and clinical understanding
of sleep patterns.

The diagram below shows a skeletal structure of our

solution.
 HARDWARE

IR CAMERA-An IR (Infrared) camera captures images in

low-light or no-light conditions by using infrared light,
which is invisible to the human eye but detectable by the
camera’s sensors. Equipped with IR LEDs, these cameras
illuminate dark areas and display images, typically in black
and white. Widely used for 24/7 security, wildlife
monitoring, and military operations, IR cameras are valued
for their ability to work discreetly in complete darkness and
various weather conditions.
RASPI 4-The Raspberry Pi 4 is a compact, powerful
single-board computer designed for a wide range of projects.
It features a quad-core ARM Cortex-A72 processor, up to
8GB of RAM, dual micro-HDMI ports supporting 4K
displays, USB 3.0 ports, and Gigabit Ethernet. With built-in
Wi-Fi and Bluetooth, it is ideal for projects like coding,
robotics, IoT, media centers, and more. The Raspberry Pi 4
offers improved performance over its predecessors, making
it a versatile and affordable tool for hobbyists, educators,
and developers.

11x20 PRESSURE MAT-A 11x20 pressure sensor mat used

for sleep recognition is a high-resolution matrix of pressure
sensors designed to monitor a person's sleeping patterns. The
mat consists of 20 rows and 11 columns of sensors, creating
a grid with 220 sensing points. Each point detects pressure
variations, allowing the system to measure body posture,
movement, and weight distribution in real-time
MICROPHONE-A microphone converts sound waves
into electrical signals using a vibrating diaphragm and a
transducer. Types include dynamic, condenser, ribbon, and
electret, each suited for applications like live performances,
studio recording, and devices. Used in communication,
entertainment, and voice recognition, microphones are
essential for capturing and processing audio effectively.

MSP430G2553 chip-The MSP430G2553 is a low-power,

16-bit microcontroller from Texas Instruments. It features a
16MHz CPU, up to 16KB Flash memory, 512B RAM, and
peripherals like UART, SPI/I2C, and 10-bit ADC. Ideal for
embedded applications, it supports energy-efficient designs
in IoT, sensors, and portable devices with ultra-low power
modes.
 CONCLUSION
The proposed pressure-sensor mattress, augmented by a
strategically placed Raspberry Pi camera, offers a
non-invasive and unobtrusive solution for long-term sleep
monitoring. This innovative approach eliminates the need
for uncomfortable wearable devices, making it suitable for
extended periods of use. By continuously monitoring
pressure distribution across the sleep surface, the mattress
can accurately detect changes in sleep posture and
movement patterns. Simultaneously, the Pi camera,
optimized for low-light conditions, captures subtle postural
details that may be missed by pressure sensors alone.
Future developments could further enhance the clinical
utility of this technology. Integrating additional sensors,
such as those that measure heart rate, respiratory rate, or
blood oxygen levels, would provide a more comprehensive
assessment of sleep health. Additionally, incorporating
polysomnography (PSG)-like functionalities, which involve
monitoring brainwaves, eye movements, and muscle
activity, could enable detailed analysis of sleep stages and
disorders.