PostureMaxxer

Inspiration

We wanted to make an app that helps correct posture because poor posture is a common problem for people in the IT/tech world who spend hours sitting at desks.

What it does

PostureMaxxer uses MediaPipe to track body landmarks and a machine learning model to classify sitting posture as “good” or “bad” in real-time, offering instant visual feedback.

How we built it

We used Python, OpenCV, and MediaPipe for real-time webcam pose tracking. We collected our own posture data, trained a decision tree classifier with scikit-learn, and built a live predictor with a GUI.

Challenges we ran into

• Getting accurate landmarks with oversized/loose clothing
• Webcam compatibility issues
• The model rapidly switching between “good” and “bad” posture due to minor fluctuations
• Learning and debugging MediaPipe and OpenCV as beginners

Accomplishments that we're proud of

• Real-time posture classification working with live video
• Collecting and labeling our own dataset
• Building a working ML pipeline from scratch
• Clean visualization with landmark overlays and angle tracking

What we learned

• How to build and train ML models from raw data
• How to use MediaPipe and OpenCV for pose detection
• How to clean noisy input for more stable classification
• Basics of computer vision and landmark-based feature extraction

What's next for PostureMaxxer

• Add calibration for personalized posture detection
• Use smoothing/averaging to reduce flickering between classes
• Make a desktop app with alerts/reminders
• Possibly build a mobile version with posture correction tips

Built With

• mediapipe
• python
• scikit-learn