Maksim Shabunin edited this page Jul 29, 2015 · 1 revision
Clone this wiki locally

RGBD (Color data with depth processing)

This page discusses what utilities OpenCV should have for handling dense depth data


  • convert a uint16 Kinect depth image to the float version (divide by 1000 and replace 0/max to NaN)
  • go from a depth image to a structured set of points using K (deals with mask and OutputArray). We have code for that in ecto_opencv
  • 3d visualizer


  • Fast normals/curvature calculation. Ethan has this, completely real time even on a CPU
  • Compute R,t, Rodriques, Quarternion to rotate one vector in 3D so that all points in one scene may be rotated to another.
    • All these forms should be easily convertible from/to each other.


  • Plane segmentation. Kurt on this.
  • Template on surface
    • First, just take a 2D template and project what it would look like on a surface in a scene, assuming that place of projection is planar
      • This is to perspectively warp, say, templates for correlation based mapping
      • Next, map the template to the actual surface
    • warp a 3d scene from one view point to another as done by Maria