3

I am interested in obtaining a 2D representation of a point cloud mapped from a known view point from which a registered optical image of the scene was taken using a digital camera. I have found an example of this here: http://live.ece.utexas.edu/research/3dnss/live_color_plus_3d.html which was achieved using Opencv (reprojected with a pinhole camera model using uscalib3D as far as I can tell), though I did not get a response from the authors as to how this was done.

I want the output range map raster to be of the same resolution in xy / same field of view as the registered optical image so that I can associate the pixel coordinates of the optical image to the xyz coordinates of the point cloud (having a raster which stores the xyz coordinates of the point cloud would therefore be more useful that a range map storing distance to the sensor). Given that the point cloud was collected from a single laser scan position located ~20 cm below the camera, we can assume that there are no occluded points from the view of the camera. The point density of the point cloud will be lower than the pixel spacing of the registered image so some smoothing of the range map must be assumed.

The inputs are:

1) 4x4 camera pose matrix

2) mx3 matrix of x,y,z coordinates of the point cloud

3) undistorted image taken from the viewpoint described in (1)

I mainly code using Matlab, so I would prefer that the solution is achievable in this programming environment (opencv mex etc should be ok). Any advice and/or suggestions would be greatly appreciated.

Thomas

Thomas Seers
  • 53
  • 5
  • Hey dude... the example you link contains Comic Sans. Yuck :-D – Luis Mendo Oct 16 '13 at 14:02
  • sounds more like [pcl](http://docs.pointclouds.org/trunk/group__range__image.html) than opencv to me – berak Oct 16 '13 at 14:15
  • @Berak: I have looked into PCL which has the functionality to build range maps, though the authors state that they used opencv to get the images seen on the link. PCL definitely has this functionality, though Iam very unfamiliar with C++ which puts me off a bit (see: http://pointclouds.org/documentation/tutorials/range_image_creation.php). In this example, I can see the introduction of a rather confusing variable 'camera pose', but cannot determine how the range image would be constrained to the FOV of the optical image. – Thomas Seers Oct 16 '13 at 16:25
  • Note that there seems to be problems using PCL as mex files and I would like a solution that can be called from Matlab ideally – Thomas Seers Oct 16 '13 at 16:30
  • Any update on this? I am struggling to do the same. – Schütze Feb 19 '19 at 21:48
  • 3
    I'd say we have too little to start, we don't have 1-3 input examples, and they are not given in the example link – Yuval Harpaz Feb 24 '19 at 06:53
  • @YuvalHarpaz there are download links at the end of the page, you just have to fill a small form. – Arthur Moraes Do Lago Feb 27 '19 at 12:39

0 Answers0