Aadavn NimalthasNimalthas Imaging

AR · Research · Real-time

LiDAR 3D Hand Reconstruction

Real-time 3D hand skeleton reconstruction on-device, fusing Vision hand pose with ARKit depth through a modular RealityKit renderer.

Stack

Swift, ARKit, RealityKit, Vision

Links

Source on GitHub

On-device
No network round-trip
Real-time
Per-frame reconstruction

A real-time iOS AR pipeline that fuses Vision 2D hand keypoints with ARKit sceneDepth to lift joints into stable 3D space. Rendered with a custom bone-only skeleton renderer and a depth-based mist point cloud for spatial consistency debugging.

Results

  • Real-time 3D hand skeleton reconstruction on-device
  • Improved joint stability under movement
  • Modular rendering architecture for future expansion
Reconstructed hand skeleton overlaid on the camera feed
Reconstructed hand skeleton overlaid on the camera feed
Depth-based mist point cloud used to check spatial consistency
Depth-based mist point cloud used to check spatial consistency

Other work

Back to software