TutorAI

An easy to use state-of-the-art AI platform for learning physical education.

Comment 1

Inspiration

Physical education is a core part of our lives that we usually tend to neglect whereas Academic studies are given sole importance. Domains like dance, martial arts, yoga often require personal tutor guidance but classes like these can be costly. Moreover, youtube is the biggest platform for these activities but does it provide security in learning? Every second person reaches out to youtube aiming to learn something new but simply watching a video is not learning. There is no assessment, no realization, and lastly no feedback for correction.

What it does

TutorAI is an educational platform and Web App that uses state of the art AI technology to provide personal and effective training in the field of Physical Education. It aims to revolutionalize the quality of learning dance, yoga, martial arts, etc that require precision in the body posture. The main objective is to create an online personal trainer and a common learning platform to bridge the gap between the learners and the trainers.

How we built it

A user-friendly Web App is built that accepts a training video be it a YouTube link or recorded video from the tutor as input. This personalized Application maintains daily progress, pose analysis, and scores of the user and ranks them accordingly into Beginner, Intermediate, and Expert based on how their performance is. Depending on the video, it provides a set of initial instructions to the user which is obtained using web scraping. Every activity is defined by a set of postures obtained from a Pose Net and segmentations. A list of pre-recorded activities is stored in the Firebase Real-Time Database.

On starting the exercise, the Website accesses the webcam and shows the motion of the user, superimposed on the original motion. This system provides visual feedback in 2D/3D for better learning. After the exercise is over, verbal feedback is provided to the user depending on how well they performed the exercise, their rank, and age group. Age group is added to distinguish the learning ability of different classes of society. A time-series comparison using DTW is done to obtain similarity between the original exercise and the one performed by the user. To consider the difficulty level of routines based on individuals a personal LSTM model is developed using federated learning to analyze movement and give quality feedback. This application also contains a feature by which people can collaborate with friends and perform activities together for a more enhanced, interactive, and fun experience.

Challenges we ran into

During the implementation of our idea, we faced issues tackling technology and computation bounds. The integration of our code also resulted in some errors in the way:

1.Reducing the time complexity of Pose Comparison. 2.Integrating the AI models to work hand-in-hand 3.Integrating the frontend and backend 4.Providing customized feedback for each user.

Accomplishments that we're proud of

The most important accomplishment that we are proud of is the integration of the various AI and algorithmic parts into a single prototype containing both frontend and backend. Some other challenges we overcame were the optimization of the AI to reduce lag and data size, implementation of various real-time graphs like gauge and spyder chart.

What we learned

There are a lot of new things that we learned in this entire hack duration like: --> Using multiple deep learning models and combining them together to a single platform. --> Connecting firebase to live website --> Complete integration of frontend and backend using flask --> Plotting complex graphs using plotly

What's next for TutorAI

We aim to build an application where more people learn things that they think cannot be done without trainers by providing a personal trainer in their hands and take this project on a large scale where this application will not just be a learning platform but also a place where people can spend their quality time in other interactive and healthy activities. We want to build a social training platform where people can connect with each other and a feed system where people can write blogs and articles on health, exercises, and fitness and also associate NGOs and awareness camps that work towards these activities. We also plan to integrate better feedback technologies for more personalization.

+ 57 more
Share this project:

Updates