HandsUp

Inspiration

Boxing is one of the most effective full-body workouts, but proper form is essential for maximizing performance and preventing injuries. However, professional coaching can be expensive, making quality training inaccessible to many. Our goal was to develop an affordable, AI-powered solution that helps boxers refine their technique and defensive skills anytime, anywhere.

What it does

HandsUp is a web app that uses computer vision and AI to analyze boxing form. It detects when a user's hands are too low—indicating weak defense—and provides real-time feedback. Future expansions will include foot positioning detection, AI-generated coaching, and velocity tracking for punch effectiveness.

How we built it

Frontend: Vite + React (TypeScript) Backend: Flask (Python) Computer Vision & AI: OpenCV, MediaPipe, OpenCV Cloud Services: AWS (Boto3), Firebase (user authentication)

Challenges we ran into

Developing an accurate computer vision model that compares the accuracy of two different videos utilizing cosine similarity.
Fine-tuning the model to accurately detect low hands in various lighting conditions
Integrating real-time, accurate AI Boxing Coach feedback without massive latency issues. ## Accomplishments that we're proud of
Developing two AI models using MediaPipe—one for detecting hand position and another for tracking punches
Implementing AWS S3 storage to automatically save images when a punch is detected
Designing a scalable system to support future features like calorie tracking and velocity detection
Crafting an intuitive and engaging user interface lucide-react and TailwindCSS.

What we learned

Optimizing AI models for real-time sports analysis The importance of user customization in fitness applications Effective cloud integration for performance and scalability

What's next for HandsUp

Expanding detection to foot positioning mistakes Adding social features like commenting and sharing workout progress Implementing a velocity-based scoring system for punches Tracking calories burned using MET calculations Refining AI-generated coaching for more personalized feedback