Patrick Schulz

Projects / Research software

KITTI scene reconstruction

Rebuilding KITTI sequences in Unreal Engine 5 with matching camera and LiDAR models.

Different reconstructed scenes from the original KITTI scenes dataset.
Different reconstructed scenes from the original KITTI scenes dataset.

A pipeline that reconstructs recorded KITTI scenes so that simulated and real sensor data can be compared directly:

  • photogrammetric reconstruction of the stereo images in Meshroom
  • fusion and colourisation of the Velodyne point clouds with Open3D
  • conversion of the tracklet annotations into world, camera and engine coordinates
  • textured meshes and a viewer with playback and annotation overlay

In Unreal Engine the scenes use the exact KITTI camera parameters, rectified and unrectified models, Brown–Conrady lens distortion as a post-process material, and aligned LiDAR and camera frames.