Spatial Computing

Mediapipe VR:
Vision-Based FBT.

A high-fidelity pipeline for single-camera full-body tracking (FBT) in spatial environments. Bridging webcam vision and SteamVR/SlimeVR backends.

BackendSlimeVR / VRChat OSC
CorePython / Mediapipe
Mediapipe VR

Hybrid Smoothing Stack

To eliminate jitter from consumer-grade webcams, the project integrates a dual-filter pipeline. The OneEuro Filter handles adaptive low-latency jitter reduction for fine movements, while Kalman Smoothing provides predictive pose stability during fast, occlusion-heavy motions.

OSC Handshake Logic

The system utilizes a custom OSC (Open Sound Control) backend to relay 16-point skeletal data to the VRChat and SlimeVR server instances. This allows for seamless "Full Body" immersion without the need for expensive IMU-based hardware trackers.

Real-Time Calibration

A Python-driven GUI allows users to perform three-point visual calibration in seconds. The system maps the detected depth from Mediapipe's Pose landmarking onto a virtual 1:1 human rig, synchronized via SteamVR.

Capabilities

  • Pose Stability: Automated jitter rejection for leg and foot landmarks.
  • Zero-Hardware: High-fidelity tracking achieved using standard RGB webcams.

Technical Benchmarks

16
Joint Trackers
60FPS
Inference Target

Performance optimized for real-time skeletal mapping with zero-latency OSC streaming.