Motion Sword
A real-time game that turns any webcam into a controller using pose estimation and low-latency motion tracking.
Building a vision controller around speed, precision and play.
The challenge
Motion control shouldn't feel like waiting for the screen.
Webcam tracking usually adds lag, loses the hands in fast swings and breaks when the lighting changes.
The goal was to build a tracker that runs in real time on a laptop and keeps the sword in sync with the player.
Understanding the data
Less latency. Better control.
The pipeline was tuned around the fast moments where players need the most precision.
Instead of tracking every joint equally, the model focuses on the wrists, arms and sword angle.
The model
Real-time vision for a physical game.
A lightweight pose model runs in the browser; a Kalman filter smooths the keypoints to keep movement stable and responsive.
Gesture detection was built on velocity and angle features, tested across players, rooms and lighting.
The model
Real-time vision for a physical game.
A lightweight pose model runs in the browser; a Kalman filter smooths the keypoints to keep movement stable and responsive.
Gesture detection was built on velocity and angle features, tested across players, rooms and lighting.
Pipeline
Fast
by design.
The final game brings
pose estimation, motion smoothing
and gesture recognition together
into one playable experience.
The main engineering rule remained
consistent throughout the project:
cut latency without cutting
accuracy.
