Im putting together a UAV for aerial surveying using stereo photogrammetry.
Id like to get some input for the experts before going ahead.
My plan is to buy a high wing trainer coupled with an APM2.
Electric powered with a gyro camera mount.
The software Im using is photomodeler scanner. It can do DSM.
Im quite a n00b when it comes to fixed wing, but I have got a fair amount of expericence on multi-rotors.
Im doing this on a fairly tight budget and need to know a good camera to use.
Any advice would be greatly appreciated and will gladly buy any1 a beer if you're ever in South Africa :)
Hi There. Yes, the whole topographical survey world is going to take a new direction once UAV photogrammetry use becomes the norm!
We have just completed a mammoth calibration test program to be able to prove our processes and accuracy of results. In general DEM surveys accuracies do not have to be down to 3 or 4 cm ( horizontal), such as for broader town planning etc. But the mines here want much better than that for stockpile volume measurements. So we executed a program where we photographed a 2.5km X 1.2km area, with a 21mm and a 30mm ( approx equivalent 35mm film camera focal length or 30mm and 45mm respectively). We photographed at 2cm GSD, 4cm GSD and 7CM GSD. We placed 140 ground tie points, 40 of them on the periphery. When we processed the images we used 30 of the periphery points and 50 of the 100 points 'randomly' spread over the full internal area. The area we used had elevation of approx 670m ASL at the North Western corner, 635m at the North Eastern corner, 521m at the South Eastern corner, and 610m at the South Western corner.So, a sort of rectangle with the SE corner draping down quite a bit ( 150meters or so) .
After processing the images and creating the DTM's/contours, etc, we then referenced the results in two ways - first we used all the balance of the placed tie points to measure and test the DTM against.
Then we identified on the DTM some potentially difficult site locations with good visual markers ( 126 new points!!) , ( embankments down to a river area, hillocks, granite ridges, etc) , took XYZ data of these from the DTM, went back to the site and located the visual markers and surveyed the points and referenced them against the DTM data.
The results are complex - not easily explained in a paragraph here! , but in essence -
Where the contours do not rise or fall more than 1meter in every 2meter horizontal distance, our average accuracy is 2.7cm horizontal, and 3.9 to 4.7cm vertical.
In instances where the rise and fall is more abrupt, for example, the sides of a river bed, say dropping 2meters within a 1meter horizontal distance, our average accuracy with 2cm GSD is 3.8cm horizontal, and 4.8 to 5.5cm vertically. ( the worst we saw is 5.7cm horizontal, 7.3cm vertical
At 2cm GSD the horizontal ground point 'creep' over 500meters is less than 5cm. This is the DTM horizontal distance versus actual measured distance over 500meters with NO tie points located between the two measurement points, and with tie points adjacent to this line, not located closer than 100meters to the line between the points - difficult to explain in words...This creep has a direct relationship between number of tie points and lens distortions. Increasing the number of tie points eliminates the effect..
Hope I never need to do this again - took a month of Sundays!
@Joe, Could you please tell me how you set and control the Camera (SAMSUNG NX200/210)?. What difference between the Samsung lens and Pentax. I think it has 16mm in the kit.
I want to know this too.
I think there are three important things that must be together to choose the best camera:
weight, sensor size (22 x 14 mm or higher) and barrel distortion.
To see distortion check
To more accurate we need the tangential distortion but it isn´t on the database and I think is too hig
About the NX200, did you get this info (what to change so you can trigger it via APM) on the net? Or did you tried and learned how to modify NX200? I'm using here in my country a SX260 + GPS, the nature of my work is much simpler than yours and high precision is not a priority, but high resolution is. I'm using SX260 because its GPS and it really helps, and I'm using Agisoft too, and I really likes it.
But if you can share I really want to know how you did it, or at least point me some site...
What used to get GPS accuracies as good with Agisoft perhaps RTK or differential?
How about, integrating the universal IR remote shutter camera control libray in the MavLink library so different type of cameras can be triggered with an IR LED from one of the I/O pins of the 2.5 board (apologies for double posting this).
Any ideas of where to start?
Let me recommend you to try to Olympus OMD or the lighter and more recent EPL5 and EPM2
the NX looks finer in terms of pixel count but it does not resolve more detail, so that just makes raw shooting imposible and further processing and storage over-weighted.
Do not assume that voightlander's and legacy jewel-glass by Zeiss and Leica must deliver better images than plasticy lenses you can find in low-budget shelves. Those lens are designed for film and often they do not perform near as good with digital sensors (this has to do with the angle of light when hits the sensor near the corners) This becomes specially true as focal-length shortens.
the 4/3 format is usually despised for being smaller than APS-C, but this makes it easier to achieve optical designs about to be telecentric.
You have a wider range of prime lenses that are designed specifically for digital sensors in this format.
Oly ZD12mm (superb!!!)
14mm Panaleica (good!)
Oly "cap-lens" 15mm (argueably the lightest and cheapest lens ever made in the world, F8 but useful)
Oly 17 f2.8 pancake (light and cheap, just decent)
New Oly 17 f1.2 mm (promising but not tested yet)
Sigma 19mm 2.8
Panaleica 20 (very good)
Panaleica 25 (wonderful but starts to be too long -50mm equiv)
Lens reset during camera switch can be overridden by setting, but it is not that much an issue as you can self calibrate in most cases.
Have you ever used a Canon camera? I use a point and shoot with CHDK sending a signal from rx to camera via USB (no servo needed) with pretty good results control-wise. The camera itself is low quality but it's all I have right now for testing. I can control it either by autopilot or rc tx and run an intervalometer script at a preset interval. I am wondering if you have any particular reasons for not using a canon such as distortion or otherwise.