Interactive Avatar with Pose Estimation
Real-time interactive digital avatars using advanced pose estimation algorithms.
Project
Real-time interactive digital avatars using advanced pose estimation algorithms.
01
Context
This project explores accessible motion-driven avatars for communication, learning and interactive applications. The objective was to animate a character in real time from an ordinary browser camera.
02
Pose-estimation pipeline
MediaPipe detects facial, body and hand landmarks from the video stream. These landmarks provide the continuous input needed to reproduce head orientation, expressions, arm movement and finger articulation.
03
Avatar animation
Kalidokit translates the detected landmarks into rotations and blend-shape parameters for a VRM model rendered with Three.js. The complete pipeline runs in JavaScript in the browser.
04
Deployment
The avatar was integrated into the GOSAI environment used by the augmented mirror. The lightweight web architecture made it reusable for sign-language demonstrations, games and remote communication prototypes.
