I am currently writing a background section on lidar sensing for part of my dissertation. I have either confused myself or not thought on the topic enough and I am attempting to gain a clearer understanding of how, and the form of, data from a lidar camera is produced.

At surface level or in terse summaries of lidar sensing you may frequently see something akin to "lidar sensors produce point cloud data". Though through my research I don't necessarily directly see how this is achieved. For example, in Paul McManamon's work "Field Guide to Lidar" he provides a high level categorical description of lidar sensors

Range-only lidar, or 1D lidar, measures only one dimension: range. 2D lidar is similar to passive imaging in that it measures both azimuth and elevation but does not measure range. A common type of lidar is the 3D lidar, measuring angle/angle/range. A 3D image can be only 3D, with no grayscale, or can measure grayscale. Polarization and color(wavelength) are other dimensions that can be measured.

Based on my experience, and McManamon's description of 3D lidar, I have seen two distinct forms of lidar imagery. The first is a point cloud, a set/bag of points in 3D space with no regular structure, grid, or ordering. The second may be described as a "2D depth image". Here the $X, Y$ values fall on a regular grid or structure and have some ordering (almost directly analogous to an RGB image $X, Y$ values) and the $Z$ values are naturally understood and viewed as greyscale pixel values where the grey intensity relates to depth or range from the viewing location.

Thus from my understanding lidar cameras do not directly capture and produce point cloud data, they capture "2D depth images". These images are then translated to some global coordinate space and co-registered. After this process the regular structure, grid, or ordering disappears as the single collections, specifically pixels, are mixed into a true point cloud.

Is this the correct understanding of how lidar sensors produce point cloud data, or is there some form of sensor which natively produces a point cloud from a single image collection?

Hopefully this is a good forum for this question, other sister sites such as Digital Signal Processing and Electrical Engineering seem to specific for this question.

  • $\begingroup$ Why do you make the assumption that all systems work identically? Anyway pointcoud data is ofter rather cumbersome to deal with z-depth is quite useful for many applications. Its still sort of point coud data just easier to parse for many usecases. Its sort of like a transpose where the points are screen oriented, instead of a possibly random point order. $\endgroup$ – joojaa Nov 7 '20 at 1:58

Are you asking how a scanning LIDAR physically works? Or just the format in which it outputs data? I feel that you're overthinking something.

Because the raw data that is obtained by a LIDAR is the angle at which the beam is pointing and the distance that was measured when pointing in that direction. It is just a rangefinder at heart after all and can only measure things relative to itself.

The way the data is actually output for the user can be in whatever format manufacturer of the LIDAR wants. Personally, I would expect the user output of a LIDAR module to be in polar coordinates, though I could understand if the extra step was taken to convert that to cartesian coordinates, your so-called "2D depth image".

If it was on a coordinate measuring machine (CMM), then I would expect the CMM to do some post-processing to convert the output data of the LIDAR module into something more akin to your "point cloud data".

I definitely would not expect point cloud data directly from a scanning LIDAR module. It just requires too much very application-specific information and post processing than would be reasonable to include on a mass-produced LIDAR scanner

  • $\begingroup$ Not how it physically works, but the final data format. At the most basic level, a collection event would be the collection and measurement of the time-of-flight of a laser pulse (plus whatever other information the system may collect from the incident light). Though in common terms, from my experience, a lidar image would be represented as the described "2D depth image". That is, a full rasterization of the laser across the scene. In this format a "true" point cloud is not produced. There is still structure and ordering of pixels related to the discrete steps in laser rotation and pivot angle $\endgroup$ – KDecker Nov 6 '20 at 21:08
  • $\begingroup$ My question is do lidar systems natively produce point clouds from the mechanics of the imaging system, or are they only produced after the fact during some post-processing? $\endgroup$ – KDecker Nov 6 '20 at 21:09
  • $\begingroup$ So by point cloud, you mean like what I would produce if I wanted to portray to you some imaginary object or building in my mind and just arbitrarily placed a bunch of points in space? Independent of perspective? $\endgroup$ – DKNguyen Nov 6 '20 at 21:09
  • $\begingroup$ Yes (en.wikipedia.org/wiki/Point_cloud). The mechanics of lidar collection impose some structure and ordering to the captured image. Whereas, point clouds do not have a structure or ordering of points, they are purely a mathematical Set of points. But, based off some literature it seems as if there is a mechanic of collection, and not post processing of the data, in which a true point cloud is produced. $\endgroup$ – KDecker Nov 6 '20 at 21:11
  • $\begingroup$ @KDecker Then no, I definitely would not expect point cloud data from a scanning LIDAR module. It just requires too much very application-specific information and post processing than would be reasonable to include on a mass-produced LIDAR scanner. $\endgroup$ – DKNguyen Nov 6 '20 at 21:12

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.