/MoveMentor

AI Instructor created at Friday Night FOSS

Primary LanguagePythonGNU General Public License v3.0GPL-3.0

INSTRUCTOR ARTIFICIEL

Overview

For the dance masters, gym instructors, football trainers and yoga gurus out there this might be unfortunate news for you cause here is a noteworthy development coming to pitch in for you. This is an AI tool that would help the user in mimicking body gestures of professionals. Putting it more precisely it would monitor and analyse the user's gestures, compares it with benchmark and suggests areas of improvement be it dancing, workout, kicking a ball, swinging a bat etc.

User would be rehearsing in front of the camera alonside which the demonstration of the video they provided would be playing. The demo could be split onto stages to facilitate enhanced learning. Once the stage is complete the application would display the insights learnt. It would provide an accuracy score on how identical the user's movement's are to that of the demo provided. It would also indicate instants which had maximum mismatch of gestures of the user and that of the ideal gestures which would help in rectifying errors effectively.

Screenshot 2023-08-09 at 10 23 16 PM Screenshot 2023-08-09 at 10 24 02 PM

A statistical curve would be reuturned which is indicative of how mathching one's gestures are to that of the actual video. On hovering over the points the user would be able to see the particular instant and his movements, also get information about possible corrections.
Screenshot 2023-08-11 at 10 22 56 PM

TechStack

  • Flask Library to handle backend
  • Deep learning model MoveNet build upon the VGG16 architecture to detect motion by tracking the 17 key points of human body
  • html, CSS, JavaScript for handling frontend
  • chart.js and D3 libraries to create the accuracy score distribution curve

Applications

The use cases of such a model are far fetching. A good proportion of people prefer to employ their skillsets such as dancing and working out in their own terms and premises . There is a good majority of people who are self reserved and have complexes due to which they might not want to practise these skills at a non familiar environment under some instructor who would not fully grasp their shortcomings.

The application could also be used to learn to mirror one's idol , Virat Kohli's cover drive, Rohit Sharma's pull shot, Lin Dan's smash Ronaldo's freekick etc to name a few.

The Accuracy and efficieny of a computer could not be mathced by regular human beings.