Public Lab Research note


DJI Phantom- Testing Camera and Mapknitter

by pswigert | March 03, 2015 21:59 | 1,394 views | 6 comments | #11653 | 1,394 views | 6 comments | #11653 03 Mar 21:59

As part of a research project I'm working on at UC Berkeley I'm exploring opportunities for sharing and collaborating on aerial imagery, obviously something Public Lab has lots of experience with.

I'm particularly interested in how commercially available quadcopters/drones will open up opportunities for aerial imagery (nothing against the kites and balloons Public Lab supports and has used effectively, those are clearly more valuable in certain use cases).

Here I just was learning to fly a DJI Phantom 2 Vision+ and wanted to test out how easy it was to take downward facing images, what quality they would be and then how to stitch them together via Mapknitter.

A couple lessons learned- it's easy with the Phantom to take photos that aren't 100% straight down. Need to make sure that toggle is fully adjusted. Also, the fisheye effect can get fairly pronounced; the crossing paths highlight it here. Post-processing to remove that seems pretty valuable if I really wanted a crisp set of images.


6 Comments

Could you add a secondary lightweight mini-camera that is fixed on Nadir to the bottom of the airframe just behind your primary sensor (with all pertinent CG calculations being done before a attempt to fly)? If not then: The fisheye distortion can be easily fixed in post processing. I have never flown a quadcopter (I'm a fixed wing guy) but Imagine getting perfect nadir imagery is hard when using a fisheye lens since the distortion makes it hard to judge your angle. Maybe try putting a stake or some type or geometric high contrast maker and try to get that in the middle of the FOV. A X shape or something like that would get visibly distorted as you get it away from center and straight down (at least that's my hypothesis).

Is this a question? Click here to post it to the Questions page.

Reply to this comment...


@Ecta64, thanks for your thoughts! I definitely think just being more careful with the actual operation of the knob that adjusts the camera angle (it was only my second time flying the Phantom) will help, and then post processing to remove the fisheye effect seems a more likely strategy for me than attaching a secondary camera. I'll be taking it out this weekend and will see how things look on a different landscape (perhaps with a marker as you suggest) and check out the various post processing options.

Reply to this comment...


Experience does help a lot with aerial imaging. Controlling a RC aircraft and then getting good imagery takes time to gain the required skills. Once you get some more flights in I'm certain the issues with positioning will all but disappear.

Reply to this comment...


To handle fisheye, you might play with Alex Mandel's code here: http://blog.wildintellect.com/blog/lenscorrect which, while specific to GoPro ATM, does lens correction with the LensFun database. It could be adapted, I think for the DJI Phantom 2 Vision+.

Reply to this comment...


Hugin can also batch process images for lens distortion and has a collection of lens profiles. I know @Eustatic has been using GIMP as well. I think he has a better note than this one somewhere.

Reply to this comment...


I find that picking up a cheap used cannon camera and loading CHDK works a lot better then having to remove the distortion from a "fisheye" lens (with hundreds of photos)...and if your post processing software can handle the fisheye, it typically takes a lot longer to process. If you do put another camera on it, I would suggest isolating the vibrations to the camera. There are multiple way of doing this, most gimbals use rubber dampers. Ive also seen thick gage wire and ear plugs used. Maybe you can use what is on the Phantom and pull the camera part off. If you cant isolate the vibrations, make sure your props are balanced. If you have any other questions, feel free to contact me and I can go into more depth.

Reply to this comment...


Login to comment.