< BeagleBoard‎ | GSoC
Revision as of 05:16, 31 March 2020 by Anirudh666 (talk | contribs)
Jump to: navigation, search

Proposal for Emotions Recognition in School Kids using Deep Learning and NLP with BeagleBone Black

Student: Anirudh Sivakumar


This project is currently just a proposal.


Completed all the requirements listed on the ideas page. The code for the task can be found in the Github repository here submitted through the pull request #143 generated in Github.

About you

IRC: Anirudh666
Github: Anirudh666
School: Manipal Institute of Technology, Manipal
Country: India
Primary language : English, Tamil
Typical work hours : 7AM to 9PM IST

About your project

Project name: Emotions Recognition in School Kids using Deep Learning and NLP with BeagleBone Black


School kids are often unrecognized in many of the Asian educational systems. They carry various emotions, right from parent’s fights to getting bullied. Kids often make intangible opinions about their life and future lifestyle based on experienced perception, thus impacting their carrier. The education system of many countries advice to identify such students and make a personal peer to peer counseling system. Apart from human intervention in decision making, we are looking forward to creating a trained algorithm that can identify such induced emotions and identify students who aren’t approachable nor identified by teachers.

Natural Language Processing (NLP) using deep learning algorithm has proven to be the best fit for decision making over human intelligence in speech recognition. A deep learning algorithm is always built and trained over time to make it best fit for small range applications like a mobile app or a desktop executable file.

The project deals in three stages. The first stage includes the use of beaglebone black and a microphone, which is capable enough to record interview voices of students from various schools in India, and store the data via the cloud. The second stage consists of storing data to be fed to a deep learning algorithm and keep training until it identifies as a best-fit percentage and the third stage deals with the design and development of a mobile application to be remotely used and to test the efficiency of the trained algorithm. The idea behind the project is to identify the emotions of the school kids and their social behavior. This helps the teachers to make more efficient peer to peer counseling. The outcome of the project would be long term research over speech recognition using NLP and Deep learning and help scientific society with best-fit speech samples for future R&D in economically slow countries.


Provide a development timeline with a milestone each of the 11 weeks and any pre-work. (A realistic timeline is critical to our selection process.)

Mar 31 Proposal complete, Submitted to
May 4 Proposal accepted or rejected
May 18 Pre-work complete, Coding officially begins!
May 25 Finalise Hardware and design and develop the PCB for the system, Introductory YouTube video
June 1 Setup a lab trial run with basic code
June 8 Interface system with the cloud storage
June 15 18:00 UTC Start building deep learning algorithm with MATLAB and Python and check the working of both these algorithms, Mentors and students can begin submitting Phase 1 evaluations
June 19 18:00 UTC Phase 1 Evaluation deadline
June 22 Visit nearby schools and interview the students to obtain data for training and developing the deep learning algorithm
June 29 Finish training the algorithm for the first time
July 6 Revisit the schools to collect more data and train the algorithm for the second time
July 13 18:00 UTC See the improvement in the trained algorithm with the students data, Mentors and students can begin submitting Phase 2 evaluations
July 17 18:00 UTC Phase 2 Evaluation deadline
July 20 Repeat the algorithm training process if required, depending on the performance of the system
July 27 Start looking into building the mobile application and integration of the 2 systems
August 3 Finalise the working of the mobile application and test it on a few more students through the mobile application, to check the integrity of the system, Completion YouTube video
August 10 - 17 18:00 UTC Final week: Students submit their final work product and their final mentor evaluation
August 17 - 24 18:00 UTC Mentors submit final student evaluations

Experience and approach

I have been part of 2 student project teams. I was in the electronics and control subsystem in Formula Manipal and I am currently the co-founder and electronics and propulsion head in loopMIT. I am also currently in the semi-finals of IICDC 2019 with the project E_agri, which is a smart agricultural system based on IoT. I have a lot of experience in working with electronics and embedded systems. I have also designed a Data Acquisition system using the BeagleBone Black for the Formula Manipal Electric car, which communicated through CAN protocol and acted as the master controller of the system. I am also set to present a research paper at the international conference of ACTSE 2020 and I am currently working on another research paper based on deep learning. I also did an internship at the Center for Artificial Intelligence and Robotics (Defence Research Development Organisation, India) where I worked on the development of a cost map for a hex-copter using ROS on Jetson TX1. Since I have worked with BeagleBone Black before, I will start with designing the hardware part of the circuit, and make a dedicated PCB for it. Then I will start with the coding for the deep learning algorithm, using Python.


If I do get stuck on a problem, I believe that I will first start looking into all the systems that are linked to that problem and I will read about those systems in detail and see if I can find other solutions online, so I can get back to the system with a better approach. Since there are a plethora of resources online, about the BeagleBone controllers, Neural Networks and Deep learning, I will be able to find a solution to the problem. If, after multiple attempts, if I am still not able to correct the system, I will approach one of my faculties at my university, who has worked in this field of embedded systems and deep learning.


If completed, it will help in understanding the emotions of not only students, but all individuals in all fields who are facing problems regarding mental health and are unable to get any help or be identified, and it can help them seek attention. I really hope I can help at least one person through this project and my project is truly complete only then. Since I will be making a mobile application out of this, it will be accessible by everyone.


The code for the task can be found in the Github repository here submitted through the pull request #143 generated in Github.


Is there anything else we should have asked you?