2026 — ongoing · Solo — CV pipeline, feature engineering, service
LiftVision
Computer vision pipeline that analyzes lifting biomechanics from video and automatically flags technique inconsistencies.
- Status
- In progress
- Stack
- Python · MediaPipe
- Areas
- Computer Vision
- 4Barbell movements analyzed
Context
Lifting technique is hard to self-assess from the inside — small form breakdowns (bar path drift, joint angle collapse) are often invisible to the lifter in the moment but show up clearly on video. LiftVision started as a way to turn raw lifting footage into structured, actionable feedback.
Problem
Build a system that can take lifting video, extract the biomechanical signal that actually matters, and flag technique issues - based on your personal body proportions - automatically rather than requiring a coach to review footage manually.
Approach
- Pose extraction. Used MediaPipe pose landmarking to track the lifter’s body across each rep.
- Feature engineering. Extracted joint angles, bar velocity progression, and bar path across 4 barbell movements.
- Rule-based flagging. Compared extracted form features against a guiding rule set to automatically flag technique inconsistencies.
- Delivery. Deployed the pipeline as a versioned Python service, including body proportion analysis to account for lifter-specific biomechanics.
Results
A working technique-analysis service that turns raw lifting video into structured joint-angle and bar-path data with automated flagging — the foundation for a broader lifting content and coaching channel currently in progress.