Thursday, December 2, 2010

Using Photogrammetry with Low Altitude Aerial Photographs

Kite Aerial Photogrammetry.
Digital Terrain Model on left and photomosaic with DTM on right.
My low altitude aerial photogrammetry work generates a lot of questions from the denizens of the internet.  Primarily, they want a detailed explanation of how 3D models and photo-mosaics are created from simple point and shoot digital cameras.  My regular work, as an archaeologist, normally keeps me too busy to answer these questions to everyone's satisfaction.  At the urging of friend, Wonmin, I have created a two-part video that demonstrates the basic steps of how I use Leica Photogrammetry Suite (LPS) with low altitude aerial photographs to create geo-referenced and rectified images. Wonmin is an extremely talented aerial photographer, whose work at a South Korean archaeological site is the subject of the tutorial.
For LPS to perform accurately, a few things are required before the aerial photographs are taken.
  1. Several Ground Control Points  (GCPs) should be placed across the area to be photographed.  GCPs are objects that can easily be seen by the camera during flight.  LPS requires a minimum of three GCPs but I have found that six or more, evenly distributed across the project area, yield the best results.  The GCPs that I often use are cheap biodegradable paper plates, but those seen in the tutorial are nice yellow and black reflective markers. 
  2. In addition to placing the GCPs, tight coordinates for each are required.  This is accomplished with a Total Data Station (TDS) or traditional transit.  Differentially-corrected GPS (DGPS) is not normally accurate enough with elevation values to be useful.
  3. After capturing several overlapping photographs of the subject matter from the sky, the photographs are ready for processing.  
  4. For processing, a calibration report for the camera is also needed. While it is possible to spend a lot of money having this done professionally, I prefer a cheaper method such as using the camera calibration function in the PhotoModeler software*.  PhotoModeler requires that several photographs of a printed grid be taken by the camera needing calibration.  The software then compares the photographs with the grid and creates a report of the exact focal length, format size, principal point, and lens distortion parameters.  These numbers can then be fed into LPS**.
Simple enough? Now, you're ready to follow along with the tutorial::
 

and Part 2
 
What's not covered in the tutorial: How to remove radial lens distortion and creating large mosaics from dozens of images.
 
* Agisoft has similar software for free but I have not tested it with LPS.
 
** It is worth noting, to spare others my frustration, that PhotoModeler measures principal point from the upper left of the lens while LPS measures it from the center.

9 comments:

  1. Great job!
    Thank you very much.
    Which version of LPS you are using?

    ReplyDelete
  2. Peter, thanks. I just wish the quality of the video had been better.

    I used LPS 9.1 but have used the same process in ERDAS 2010 without problems.

    ReplyDelete
  3. Hello Mark.
    At the momment I am obtaining my firsts results with Photomodeler, bun I find some inconveniences with it. My question is about the calibration of the camera, do you have resolved it?, because I have mine calibrated with photomodeler and I want to start using LPS.

    Thank you y Saludos!

    ReplyDelete
  4. Francisco, I'll have a tutorial on how to convert a Photomodeler camera calibration to LPS soon. -Mark

    ReplyDelete
  5. What is the difference between doing this and simply using a total station to do what seems like the same job? Especially since the total station seems to be required anyway?

    Thanks,

    Charles

    ReplyDelete
  6. Charles, thanks for checking out the blog.

    There are a couple of significant differences between using this process and just using a total station. Chief among these is time. To make a DEM of similar density, using just a total station would take forever. Weeks probably. Compare that to the few hours it took to capture the data used in the tutorial. Furthermore, the aerial photographs are also created. As an archaeologist, I find the aerial images extremely useful.

    A total station is not mandatory to this process but it does increase the accuracies. Having good Z values is most important. A traditional transit or a differentially-corrected GPS can also do the trick. An inexpensive way to setup a DGPS is outlined here: http://gpspp.sakura.ne.jp/rtklib/rtklib.htm and here http://gpspp.sakura.ne.jp/rtklib/rtklib_beagleboard.htm

    ReplyDelete
  7. Any ideas on a way to combine the DEM with the photos to create an orthophoto?

    ReplyDelete
  8. Chad, sorry it wasn't clear. The process also produces orthophotos that can be superimposed over the DEM. If you click the "Kite Aerial Photography" image at the top of the page, you'll see an enlarged version. The right hand image is a partially transparent ortho-mosaic superimposed over the custom DEM.

    Checkout the poster of the imagery here: http://www.palentier.com/Turtle_Ridge_Poster01.pdf (~4 Mb)

    -Mark

    ReplyDelete
  9. Hola Mark, thank you very much for these tutorials.

    Do you have any idea about the precision that you are getting using this method, and with the control point surveyed with a total station... I mean are you getting a model with precision at milimeters... centimeters.. can you apply this method to make a engineering details over this model?

    Cheers

    ReplyDelete