PosePilot X bitcamp@2025
General Beta interface
Alerting the user if poor posture maintained for so long

PosePilot: Real-Time Posture Monitoring Assistant

Team Members: Sai Jagadeesh Muralikrishnan, Chan, Pure, Vivek
BitCamp 2025 Submission

Inspiration

The inspiration for PosePilot came from a shared struggle that many of us face—spending long hours in front of our computers with poor posture. As students, gamers, and developers, we often found ourselves slouching unknowingly for extended periods, leading to neck strain, back pain, and fatigue. We wanted to build something practical and helpful—a lightweight tool that could nudge us toward better posture in real time, without needing any wearables or specialized hardware.

What We Built

PosePilot is a real-time AI-powered posture monitoring system that uses your webcam to detect your posture, evaluate key metrics like neck and back angles, and provide visual/audio feedback if you’ve been misaligned for too long. The system also gives you a posture score, deviation alerts, and motivational badges to gamify progress and encourage good habits.

What We Learned

We deepened our understanding of pose estimation using YOLOv8-Pose.
We applied vector math and trigonometry (like dot products and angle calculations) to convert raw keypoints into posture insights.
We learned how to build and deploy Flask APIs, integrate real-time camera feeds, and render feedback through a fully responsive frontend built with Tailwind CSS and HTML.
We explored how to deliver high FPS performance in Python while balancing accuracy with efficiency.
We also discovered how important user experience is in an AI tool—so we designed a clean, intuitive interface to make posture tracking simple and enjoyable.

How It Works

Your webcam captures real-time video frames.
The backend uses YOLOv8-Pose to extract keypoints from each frame.
Using vector angles between your shoulders, hips, and neck, we calculate:
- Neck tilt
- Spine curvature
If the neck/back angle exceeds a threshold (typically ~20°), we trigger posture alerts.
A real-time posture score updates as you maintain better posture over time.

We also included a baseline posture capture, deviation duration tracker, and earned badges like "Posture Master" and "Fast Tracker" to help users stay motivated.

Built With

Python – Backend processing and detection logic
YOLOv8-Pose – Real-time pose estimation
Flask – Lightweight API backend
HTML5 + Tailwind CSS – Modern frontend
Chart.js – Posture history graph
Pyttsx3 & Audio Alerts – For voice feedback and beeps
Three.js & GSAP – For 3D visuals and scroll animations
(Upcoming) Google Gemini AI – For posture reports and tips (in progress)

🌐 Try It Out

💻 GitHub Repository: github.com/cravotics/Pose-pilot
Follow the instructions in the README to replicate the system locally.
📸 Note: The demo requires running the Python backend (main.py) and accessing the HTML interface locally.

Challenges We Faced

Calibrating neck and back angles reliably from webcam input was tricky due to real-world camera distortions and body variations.
Managing real-time performance with high-resolution video while maintaining inference speed required careful optimization.
Designing a UI that felt natural and motivating without overwhelming the user took several iterations.

What's Next?

We're integrating Gemini AI to generate daily posture summaries and personalized feedback.
We're also prototyping a vibration wristband that gives a gentle buzz when you slouch—turning PosePilot into a full hardware+software assistant.