Project for HackJNU-2.0 (Track - Machine Learning & Health)
Pandemic has made it difficult for us to interact with our friends and enjoy some time playing or doing something competitive with them. We interact with them virtually but the activites are limited to playing some online games. As rightly observed, the physical fitness and health among WFH folks has degraded over time. With gyms, parks and other walkways following restrictions and caution, it has become difficult for health consious people to interact. And what is a gym if you can't follow up your crunches with few friends and folks competing with you.
Our solution presents a virtual gym that shall be based on pose detection. Here people can join in through a private video chat room and exercise with their mates. Our pose detection algorithm shall take care of the counts and posture of the participants. There will also be a dashboard or leaderboard where we can see our scores with other folks. Having a personal dashboard and page for solo training should also be a great addition. So one can create a session with multiple people in the room. Once the timer Starts people can follow up the instruction on the exercise they have set.
Hopefully we are looking ahead for the following social impact:
-
People can track out their health easily.
-
Following up some basic exercise should not require a big setup or spending.
-
Collaboration among peers and conducting those exercise sessions must become easier.
-
Since it involves having it with peers, we hope that youth finds it more interesting and is drawn towards it.
The tech stacks will include:
- Machine Learning
- Javascript
- Node.js
- MongoDB
- socket.io
- peer.js
- Web RTC
- mediapipe pose
- Bootstrap
- OpenCV
-
brief problem statement discussion
-
pose detection implementaion locally
-
video streaming on web
-
integrating multiple streams
-
basic website layout design
-
user authentication
-
dashboard for multiple users
-
integrating streams with pose detecting model in javascript
-
gamification of the idea
-
features like badges and incentives
-
add more functionalities and incentive based approach for gamification
-
instead of hard-coding each exercise, a ML based approach can be used for montoring the exercise being done currently. It will need a dataset with landmark point depicitng action that represents a particular pose.
-
Improving the UI/UX will also be a good todo for us, currently we bootstraped a simple UI for the demonstration and deployment purpose.