Problem

As computer vision interns at WeSkill, we were tasked with developing an AI-based dance learning application that could be used by thousands of students. The challenge was to create a platform that could accurately detect and estimate human body poses in real-time while also incorporating frame-by-frame syncing with the audio signals.

We faced several obstacles during the development process, including finding the right model for human body pose detection and estimation, managing large datasets efficiently, and ensuring seamless integration of all components of the platform.

Check out our demo video here!

Solution

After extensive research and experimentation, we decided to use the BlazePose model for human body pose detection and estimation. We also implemented MYSQL Workbench for efficient storage and access of relational data and GCP buckets for object data management.

To ensure seamless integration of all components, we used Flask as our web framework and deployed the application on Heroku. We also incorporated frame-by-frame syncing by comparing spectrograms of audio signals using the FFT method.

Impact

Over 10,000 students are currently using our AI-based dance learning application.
The platform has received overwhelmingly positive feedback from users who appreciate its accuracy in detecting and estimating human body poses in real-time.
The project has helped us develop valuable skills in computer vision, database management, web development, and deployment on cloud platforms like Heroku.

"This project has had an incredible impact on the way we approach computer vision and web development. We are proud to have created a platform that is helping thousands of students learn dance more effectively."

Team

Team Member #1 - Computer Vision Intern
- Implemented human body pose detection and estimation using BlazePose model.
- Managed large datasets efficiently using MYSQL Workbench and GCP buckets.
Team Member #2 - Web Development Intern
- Built the web framework using Flask and deployed the application on Heroku.
- Incorporated frame-by-frame syncing by comparing spectrograms of audio signals using the FFT method.

AI Dance Learning App Helps 10,000+ Students