IITGN’s Summer Research Internship Project

(SRIP Project Number: IP0NB0000020)

Table of Contents
About the Lab...............................................................................................................................2
Alumni Experience:..................................................................................................................2
Mentor:.....................................................................................................................................2
Past SRIP interns in action...................................................................................................... 2
Procedure for Application:......................................................................................................... 2
Diagnosing Sleep Apnea using Thermal Imagery.................................................................... 3
Problem Statement 1: Modeling Sleep Apnea.........................................................................3
Background........................................................................................................................3
Target................................................................................................................................. 3
Prerequisites...................................................................................................................... 3
Reading Materials.............................................................................................................. 3
Problem Statement 2: Finding the Best Camera position using Bayesian Optimization......... 4
Background........................................................................................................................4
Target................................................................................................................................. 4
Prerequisites...................................................................................................................... 4
Reading Materials.............................................................................................................. 4
Problem Statement 3: Hardware device for Screen Apnea..................................................... 5
Background........................................................................................................................5
Target................................................................................................................................. 5
Prerequisites...................................................................................................................... 5
Reading Material................................................................................................................5
Sound Bubble. Active Noise Cancellation for better sleep..................................................... 6
Background........................................................................................................................6
Target................................................................................................................................. 6
Prerequisites...................................................................................................................... 6
Reading Material................................................................................................................6
Vibration Sensing Touch Panel.................................................................................................. 7
Background........................................................................................................................7
Target................................................................................................................................. 7
Prerequisites...................................................................................................................... 7
Selection Criteria......................................................................................................................... 8
Selection Task #1.....................................................................................................................8
Selection Task #2.....................................................................................................................8
Frequently Asked Questions (FAQs)......................................................................................... 9
About the Lab

Sustainability Lab focuses on utilizing machine learning

and sensing technologies to address computational
sustainability challenges, particularly in smart buildings,
energy disaggregation, air quality, and healthcare.
[https://sustainability-lab.github.io/]

Alumni Experience:
Read about the Awards and Alumni experience.

Mentor:
Ayush Shrivastava
Email : shrivastavaayush@iitgn.ac.in , Website : https://ayushshrivstava.github.io/

Past SRIP interns in action.

●​ Devodita Chakravarty [SRIP 2024] won the Bhalodia-Khetan Summer Research
Excellence Award for the project titled “ML for Sustainability: Satellite Data
Preprocessing for Detecting Pollution Sources Using Computer Vision.’ The award
includes a cash prize of ₹50,000 and a Certificate of Excellence”

Procedure for Application:

●​ Step 1 : Please fill out the form on the official IITGN SRIP portal. Take note of your
InternID. You will need it to fill the google form attached with this document.
●​ Step 2 : Please fill the Google Form and solve any one task shared from this document.
●​ Push all the code, documents related to the task to a github repo and share using the
google form attached in this Document. Deadline : 1st March 2025
●​ If in doubt, contact the mentor (Ayush Shrivastava) to learn more about the task. All
submission needs to be made in the form of a github repo.
●​ Please read FAQs for some clarifications.
Diagnosing Sleep Apnea using Thermal Imagery

What is sleep apnea?

Sleep apnea is a sleep disorder in which breathing repeatedly stops and starts during sleep.
These pauses in breathing can last from a few seconds to minutes and may occur multiple
times per hour. It often leads to fragmented sleep and decreased oxygen levels, which can
cause daytime fatigue, cognitive impairment, and long-term health risks such as hypertension,
heart disease, and stroke. Youtube video for better understanding.

Problem Statement 1: Modeling Sleep Apnea

Background
As per the medical guidelines set by American
Academy of Sleep Medicine (AASM). An apnea
event is categorised as 80% reduction in nasal
airflow and an hypopnea event is categorized as
30% reduction in nasal airflow.

Target
Model Respiration signal time series data to
estimate apnea events.

Prerequisites
●​ Working knowledge of Python, Machine
Learning, and Sequence Modeling.
●​ Interest and openness to exploring outside the Computer Science domain.

Reading Materials
●​ Learn about Sequence models here and here
●​ An example of modeling is provided here
●​ Learn about feature engineering for Time series Data here .Use TSFEL library

Mentor:
Ayush Shrivastava [shrivastavaayush@iitgn.ac.in]
Problem Statement 2: Finding the Best Camera position using Bayesian
Optimization.

Background
Assume We have a thermal camera positioned at the edge of the bed, mounted on a motorized
framework. The camera runs an object detection model that attempts to localize the patient’s
nostrils, providing a confidence score for each prediction. Our goal is to use Bayesian
optimization to determine the optimal camera position that maximizes the model’s confidence in
detecting the nostrils.

Target
Create an end to end Raspberry pi based device that has a thermal camera attached to it. This
thermal camera takes in the Thermal feed and reaches the most optimum position for capturing
nostril respiration. The movement needs to be controlled via a motorised system.

Prerequisites
●​ Working knowledge of Python, Machine Learning, and Computer Vision techniques.
●​ Experience in OpenCV, Raspberry Pi, Bayesian techniques.
●​ Experience with Servo motors, Arduino, Raspberry pi is a must. Needs to have some
hardware knowledge.
●​ Interest and openness to exploring outside the Computer Science domain.

Reading Materials
●​ Nipun Batra’s Lecture on Bayesian Optimization
●​ Machine Learning Mastery’s youtube video on Bayesian Optimization
●​ Article on Bayesian Optimization by Aporv Agnihotri from sustainability lab
●​ Article showcasing the use of Skopt library.
●​ Roboflow’s Youtube series on using YOLOv8

Mentor:
Ayush Shrivastava [shrivastavaayush@iitgn.ac.in]
Problem Statement 3: Hardware device for Screen Apnea.

Background
Screen apnea refers to a phenomenon
where individuals unconsciously hold their
breath or breathe shallowly while using
screens for extended periods. This behavior
often occurs during activities such as
working on a computer, using a smartphone,
or watching TV. It can lead to symptoms like
fatigue, anxiety, and decreased
concentration. Screen apnea is thought to be
a response to the focus and concentration
required during screen use, causing a
disruption in regular breathing patterns. For
further reading please refer to these articles
here and here.

Target
Use a thermal camera feed which is connected and processed on Raspberry pi in real time to
localize the nostril region and then extract the Nasal respiration from the localized region. Show
the respiration signals of the user on a screen.

Prerequisites
●​ Working knowledge of Python, Machine Learning, and Computer Vision techniques.
●​ Familiarity in using OpenCV, Raspberry Pi and similar hardware devices.
●​ Interest and openness to exploring outside the Computer Science domain.

Reading Material
●​ Roboflow’s Youtube series on using YOLOv8
●​ FreeCodeCamp’s Youtube video on using OpenCV
●​ Read about Bandpass filter and other smoothing filters like Savitzky Golay filter

Mentor:
Ayush Shrivastava [shrivastavaayush@iitgn.ac.in]
Sound Bubble. Active Noise Cancellation for better sleep.

Background
Snoring can disrupt sleep for both the snorer and their
bed partner, leading to fragmented rest and daytime
fatigue. The loud, irregular noise can make it difficult for
others to fall asleep or stay asleep, often resulting in
frustration and sleep deprivation. Over time, poor sleep
quality can contribute to mood disturbances, reduced
cognitive function, and strained relationships. In severe
cases, snoring may also indicate underlying health
issues like sleep apnea, which can have more serious
consequences if left untreated.

Target
We wish to engineer sound waves from smartphones placed near a person sleeping such that
these waves cancel out the noise from the surrounding environment. Watch this video for better
understanding.

Prerequisites
●​ Working knowledge of Python, Machine Learning, and Computer Vision techniques.
●​ Familiarity with Signal Processing, Fourier Analysis, Spectrograms, FIR filters, etc.
●​ Background of signal and systems, Digital signal processing, Control system would be
beneficial.

Reading Material
●​ Youtube video on Active noise Cancellation.
●​ Youtube video on understanding Spectrograms.
●​ Learn about feature engineering for Time series Data here .Use TSFEL library

Mentor:
Ayush Shrivastava [shrivastavaayush@iitgn.ac.in]
Vibration Sensing Touch Panel

Background
Touch-sensitive surfaces are widely used in modern interfaces,
but most rely on capacitive or resistive sensing, which requires
direct electrical contact. An alternative approach is using vibration
sensing to detect touch events. When a user taps or touches a
glass panel, vibrations propagate through the material, which can
be detected using accelerometers placed at the panel’s corners.
By analyzing these vibration signals, it is possible to estimate the
touch location, creating a touch-sensitive interface without the
need for specialized coatings or embedded circuits.

Target
Develop a prototype system that converts a glass panel into a touch-sensitive interface using
four accelerometers placed at its corners. The system should:

●​ Capture vibration signals using accelerometers.

●​ Process the signals to estimate the location of touch events.
●​ Display real-time touch coordinates on a connected interface.

Prerequisites
●​ Experience with Python, Signal Processing, and Machine Learning.
●​ Familiarity with Raspberry Pi, I2C/SPI communication, and accelerometer
●​ Knowledge of time-series analysis is beneficial.

Mentor:
Ayush Shrivastava [shrivastavaayush@iitgn.ac.in]
Selection Criteria

Selection Task #1
●​ Use UCI HAR dataset and build a model to predict various activity classes.
●​ Use Deep Learning models like LSTM, 1D Cnns for modeling using inertial sensor data
(Raw accelerometer data). Do not train DL methods on already existing features
provided by the authors.
●​ How can the same be done using Machine Learning models like Random forest, SVM,
and Logistic regression? Use the TSFEL library to generate features from Inertial data.
Compare the performance of models trained on your generated features versus features
provided by the authors.
●​ Make an .ipynb file to demonstrate modeling using Deep Learning and Machine learning.
Create a short summary explaining the approach and observations.

Selection Task #2
●​ Interface an ADXL345 or LIS3DH accelerometer with Raspberry Pi or Arduino using
I2C/SPI and read X, Y, and Z data at 100 Hz.
●​ Stick the accelerometer on a glass surface and tap on a glass surface, record the
accelerometer response, and store it in a CSV file.
●​ Plot the readings, and identify spikes to analyze tap vs. no-tap states. Implement an
ML-based tap detection and print "Tap detected!" when triggered.
●​ Submit a Python script (.py or .ipynb), a CSV file, and a short summary explaining the
approach and observations. ipynb files need to be submitted using a GitHub repo.
●​ You can create a video and demo your prototype.

⭐ Brownie points for being able to solve both the selection tasks !!

Note: Individuals working in AI are well aware of the online resources available and can easily
differentiate between code generated by a language model and that written by a student. We
strongly encourage originality in your work. Any instance of plagiarism may result in the
cancellation of your candidature.
Frequently Asked Questions (FAQs)

1.​ Do I need to solve both the selection Tasks?

a.​ Only one needs to be solved.
2.​ Who should I contact when in doubt?
a.​ You should write a mail to shrivastavaayush@iitgn.ac.in keeping
nipun.batra@iitgn.ac.in in cc.
3.​ What if I am unable to complete the entire question? Do I still have a chance?
a.​ Yes, You still have a chance. However, we still encourage you to solve as much
as you can.
4.​ There are two forms. Which one should I fill first?
a.​ You should fill out the form on the IITGN SRIP portal first, obtain the internID, and
then fill out the Google form provided with this document.
5.​ Does completing the tasks ensure my acceptance?
a.​ No. Completing the task is just a prerequisite. It makes you eligible for the
opportunity.
6.​ Can I mail you my submission instead of submitting a GitHub repo?
a.​ No. We require the submission to be made as a GitHub repo. You should
mention your repo when submitting the Google form. We recommend learning to
use Git and Git Hub.
7.​ How can I obtain internID?
a.​ You get an auto-assigned intern ID when you apply on the IITGN’s SRIP portal.
Use that Intern ID to fill out the Google form.

8.​ Are there any other positions available in the lab?

a.​ Yes. Please visit the Openings page on the Sustainability Lab’s website.
9.​ My question is not here in the FAQ section. What should I do?
a.​ Read more in SRIP Guidelines. If you have more questions please contact me
via email <shrivastavaayush@iitgn.ac.in>.