The Range Image vs. The Point Cloud

Elmer Bol, Director of Reality Capture at Autodesk says he does not use the term point cloud. He prefers the concept of a range image. Elmer explains that a range image is usually a 5,000 x 10,000 matrix of points, per scan where the X and Y coordinates are stored, with range as the third coordinate value.

It helps to visualize this for a moment. Most surveyors think of Z as height/elevation, as with GNSS, but that is not the case with laser scanning. When you think about it I believe this does make more sense than the term point cloud, at least from the software side.

Your thoughts.


  • In cases of a full 3D point cloud, the term “range image” has some problems. In addition there are the range cameras, which produce depth images very much resembling what we (humans) understand by the term “image”. And what’s more the term “point cloud” definitely sounds more intriguing and romantic than range image !!

  • I suspect that if you talk to the FLASH LIDAR folks you’ll find they probably do the same thing.

  • As far as I know, there are some differences between a range image and a point cloud.

    A range image is really 2.5D as there are presumed to be no overlapping regions within the image.
    A point cloud, if it is produced from multiple scans cannot be defined by a 2.5D image as there are overlapping points. This makes it fully 3D.

    The data from a single laser scan is typically composed of a matrix grid of evenly spaced points. This grid can be viewed as a range image if there are not multiple returns per pulse… In the case of state-of-the-art lidar systems, you will find multiple returns in many cases, which would either be difficult, or not possible to display in a range image as it involves ‘stacking’ multiple ranges within a single pixel.

    Another thought is that a range image is typically (again, this is based on my experience) based on a a square pixel, such as provided by a camera. A laser’s cross section (aka spot) is typically not square like a pixel is. It is typically some circular shape, and therefore deriving a range image from a scan involves some morphing of the points’ shape to fit the gridded raster of images.

    • I like your explanation Ananda. I could detect some inferences to photography with the term Range Image but you helped clarify it.

      Maybe it’s just me but I have a problem with people trying to invent their own terminology for things that have become readily adopted and accepted, like the term point cloud.

      Also, I could see some conflicts developing with two of the items mentioned: First, the 5,000 x 10,000 matrix. Why would this be a limited matrix? I think this shows it’s roots came from photography. Second, thinking of range as the 3rd coordinate value – in some CAD programs, X is a horizontal axis with positive values to the right, Y is a horizontal axis in and out of the screen, and Z is height or elevation. With Elmer’s method, Y and Z get transposed.


  • This candle-light-coder agrees: “point cloud is more romantic than range image”. Also if you use range image then you have excluded multi-return LiDAR systems. For a range image you are only allowed one “z” value per laser pulse. So you would have to settle for the first-return or last-return cloud to call it a range image.

  • I think range image is a pretty good name.

    Usually you have better range precision close to the sensor. also the angular precision is probably different than the range precision. This gives an point position error that is anisotropic. A point cloud to me is a derived model whereas the range image is more like the original data. There is no problem to store multiple hits in one pixel (rixel anyone?). There are a number of rich pixel formats that can handle that. look to the vfx 3d rendering formats like EXR and perhaps RPF.

    Also you might have a range image that has an unknown field of view that needs a specific “projector” to register into a true 3d point cloud.


Leave a Comment

This site uses Akismet to reduce spam. Learn how your comment data is processed.

%d bloggers like this: