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 :)
Good to hear someone else using photomodeler scanner. From my research, this is the software I'll be using too. I plan to use ground control points to give scale to the project and improve accuracy. Eventually I hope to use hardware to give precise RTK GPS coordinates of the camera for each picture, potentially negating the need for ground control points. My goal is to make topographic surveys accurate enough for engineer design use. I have not picked a best camera yet, but it will be a mirrorless non-zoom lens camera. Sony is coming out with a new "RX-1" fixed lens mirrorless full frame 24mp camera, but it's not cheap, $2,600. I'm budgeting for a $600 model. I read somewhere else on here that CMOS chip cameras should be avoided, that they do not take good pictures because of the vibration. I do not know if this is true. Most all cameras use this type of sensor. Good luck with your progress. I'll be looking at the forum for all photogrammetry related posts.
Dont waste your time with Photomodeler....Have tried a number of softwares and to date the best performer is Agisoft ( google it) It works exceptionally well and generated full stitched images with very good DTM's and fully orthorectified imagery.
Generates very good outputs for survey grade data and works well for mine applications as well, eg, stockpile volume determinations, etc.
Check my posts on SurVoyeur - we have now flown more than 900linear km of area and processed all the imagery with Agisoft, with survey grade results ( better than 4cm horizontal ground accuracy and 7cm vertical accuracies), and satisfed customers!
Not sure what other posters are implying re CMOS sensors - nothing to do with sensor type as far as vibration, etc. For the best results:
At least a 18Mpixel camera. Samsung NX200 or NX210 is what we use - 20Mpixel, with Pentax 21mm Prime lens - one of the best performers with regard to barrel distortions. 21mm allows reasonable flight heights for good ground resolutions. We are currently evaluating the Pentax 15mm prime lens as well, which may be even better suited. However, I suspect the VoigtLander 15mm lense might be the real solution, it is a bit more than half the weight of the Pentax 15mm lense, with the same performance, it claims.
The Sony NEX seies are more popular than the Samsung NX range, but the NX5 is only 16Mpixel and that is NOT good enough for when used with the 15mm lens - 15mm lenses allow much greate area coverage for a given height, and coupled with high pixels sensors allows down to 4cm ground resolution with ease. The Sony NEX7 is 22Mpixel, BIT is 120grams heavier than the Samsung. Also it is about 1.7times the price!!
Support from Agisoft ( Russian) is VERY good. We fly with 60% forward overlap and 60% side to side overlap in imagery - implies lots of photos to process, but you need that to be able to generate the accuracies in ground resolution needed for survey grade data. Don't let anyone tell you otherwise - if they do , they have never put there data to a true accuracy test!
It is also useful to stablise the camera in the roll axis - any light weight fixed wing plane is more agile in roll than pitch, and this helps the photograph sightline, although this can be compensated for by flying with greater side overlap = just means longer image processing times. We also do road and pipeline surveying, and a roll stabilised camera is very usefull here - the roads and pipelines have 'kinks' that require the A/C to roll - so camera stabilisation helps.
Most important for data accuracy. however, is that you have MANY tie points, ie, photographically visible target, accurately placed with survey grade GPS systems. These are then used by the processing software to 'morph. the skewed photographs into accurate orthorectified data. The more tie points the better.
Hope ths helps..
Joe, The NamPilot, Namibia..
Thank you very much for the very informative message! Very cool to hear that you are doing what I would love to be doing. Here in the states as you may know, we have to wait until 2015 when the FAA will supposedly legalize commercial use of UAV's. I'm fine with that, it will give me a few years to learn everything in my spare time as a serious hobby. I'm a licensed surveyor and I see this as being a game changer for the way surveyors create topographic surveys. No longer would we need to hire the services of a photogrammetrist using a Cessna Airplane with a million dollar camera, nor would we need to field survey with GPS or a total station to locate thousands of points. (except for the obscure areas covered with trees.)
I have read many good things about Agisoft, and I would consider them, but the main reason I may go with Photomodeler scanner software is because I have their older software from many years ago, and I could get the Photomodeler Scanner for the upgrade price of $1,595 instead of the full price of $2,595, or the price of Agisoft for $3,499. I have heard bad things Photomodeler scanner crashing, but after looking into it, I found out this was with their older software running on a 32 bit system. I'm told that it can now handle hundreds of pictures running on a 64 bit system.
In your tests with Photomodeler, did you possibly discover discrepancies with accuracies?? I'm curious what you did not like about Photomodeler or if it was maybe simply the case of Agisoft being better. I'm thinking I may need to consider spending the extra money for Agisoft.
I also looked into Leica Photogrammetry Suite. I imagine that would do a fine job too, but it was $13,398! A bit too steep for me.
Have you ever looked into a system called "V-Map" from microaerialprojects dot com? Their system uses a GPS base station and rover antenna on the UAV to record the RTK GPS coordinates of the camera right when the picture was taken. I'm not sure if it would totally negate the need for ground points, but supposedly it could.
I looked at your SurVoyeur project, wow! you really are into this. I wish you the best of luck and hope you can make an international business out of this.
Thank you once again,
Thank you for the information. I'm just start to use APM for aeromapping. Regarding the Agisoft, what is the requirement/specification of the computer? I've read somewhere it require 256 GB of RAM, is it true? Thank you
Hi There Ty,
Sorry for the delay in responding - life takes its toll...
I am not a surveyor, but most of what we do in the UAV side is for surveyors, so we are regularly involved in training the people on our system, and sorting out the problems with image processing. As a by product we have had to learn to use a image processing package, and spent many many hours doing so - Photomodeler, Agisoft, etc. We did also try very hard to get the Leica Photo Suite to work - we had a 4month demo package with full support from Leica, who eventually gave up, siting 'poor aircraft stability' as the reason there software would not generate a workable result! There system does want A/C attitude data to aid in the image center determination, and places a large emphasis on its accuracy - which you will not get from cheap mems gyros aloft in a sub 4kg plane being tossed about in the upper air thermals...
We also tried the Adamtech ( Ausy) software which also worked quite well, but suffered severe overload when fed more than 60 or so photos...
Agisoft gives very good results. For example, if we fly a 200hectare area, we get around 900 photos ( we use a forward overlap of approx 70% and a side to side overlap of approx 60%). We would fly at a height of around 140meters AGL for a GSD of 2cm. These images would take around 75hours to process on Agisoft, with a tested GSD of 3cm horizontal and 4.5 to 5cm vertically When given the logged A/C attitudes and projected GPS photo centers, the processing took 63hours. The outputs were a fully orthorectified image, the DTM and contours at 8cm intervals. The number of GPS tie points we would use for this would be around 20minimum, preferably 30 to 35, say 15 to 20 around the periphery of a polygonal area, and the balance internally.
Photomodeler generated the stitched image 10 times faster than Agisoft, but never managed to create the ortho..It managed when fed images in chunks of less than 100, but took more than 40 hours per chunk, and when trying to feed it processed chunks, it also was unable to process after given 3 of 100 photo chunks -- gave up at that point! The Photomod. user interface is much nicer than Agisoft, lots of neat features and buttons, but it does need to deliver the goods....
Do NOT believe any system that says you will not need ground GPS tie points! As you do not have the 'million dollar' camera and IMU on board, the only way to accurately process the images for perspective distortion and progressive distortion creep over the area, as well as for lens distortions, is to have many tie points to force the ortho fit into place.
We record the A/C GPS position when the photo is taken ( we modify the camera to be controlled by the autopilot) and we record the pitch and roll angles of the camera ( we also have now a roll stabilised camera) and then we also compute the image center ground co-ordinates for that image - all this within the accuracies of the MEMS IMU...
Having a very accurate co-ord of the exact aircraft position the photo was taken at is only a small part of the solution. If your attitude data is not equivalently accurate, the problem is not solved. There is a GPS system which may help - the Septentrio PolaRx2@ or its latest successor. This is a system with 3 or more antenna that have to be quite widely spaced on the aircraft - you then get very accurate attitude and position data - supposedly Survey grade position accuracy, and down to 0.2deg attitude accuracy - I have acquired such a system and will be testing it soonish - But I do not believe it can be that accurate in a moving platform, etc. We will see...One problem is that the antenna need to be far apart - better than 2 meters - and that will not work on a small UAV - maybe one on each wing tip of a 2m wingspan, but the third antenna ?? maybe a very long tail...And the unit is not light weight - around 1kg for unit plus 3 antenna...
Agisoft needs none of this - it just works....
A useable system would be:
The FASTEST quad core processor you can get hold of.
at least 16GB Ram, prefer 32GB ( 256GB??? not required!)
A VERY fast and large hard disk ( 1 terraByte...)
If not a fast hard disk, a useful addition is a large fast RAM drive
WIndows 7 unfortunately... ( must be 64 bit system...)
The best Graphics card you can get
Very Important - A UPS - very frustrating to lose 60 hours of processing because the power failed...
Hope that helps
Very interesting post, I like meet people that can compare different softwares, I want buy one software for microtopography, I'm looking a package and I like Agisoft (cheapest) but I see other like Ensomosaic more expensive but seem more profesional and have options for multiespectral and image enhancement.
I'd like too much know you valoration.
I've done a little bit of background work in the evaluation of DEM accuracy for small UAV missions.
Have a look at my 2012 research on UAV derived DEMs.
Thanks for your info. I may want to try droneMapper since their price looks reasonable. The system you suggest may cost over 1500 USD and the software cost over 3000 USD. Not that economical for small project or like myself just start to get involve in UAV aeromapping. Thanks for your very informative sharing.
Thank you so much for the very informative message. Your message holds so much good information for anyone in the future doing a word search looking for information on the different software packages mentioned.
Looks like I need to seriously consider the Agisoft software.
That's some very good points you make on the necessity of needing ground control points. We've always used them in surveying when we give the photogrammetrist the information they need. We tend to use a photogrammetrist if the topo is much more than 10 acres, and we never do much more than a hundred acres. I'm in a civil engineer company, so we would not have a need for anything much larger.
Hey, I saw this in one of my emails, something you might possibly be interested in and could possibly qualify for. It is an international conference, they are looking for presenters. www (dot) sparpointgroup (dot) com/international/ On the top right there is the link of "Call For Presentations - Deadline October 31!" I have no idea if they pay for the presenters traveling expenses, but if so, it may be something for you. You have a unique situation in that you are doing UAV photogrammetry commercially (not yet allowed in the states) and that you are doing it on a very small scale budget with your own planes.
I wish you the best with your UAV photogrammetry work, it is an industry that I think is on the edge of becoming big.
Great input!!! I have not tested Agisoft, but have utilized PhotoModeler with some great results for our mapping applications. I'm not saying it's the best, but it has some really good tools for our uses. I would be interested to learn what accuracy processes you utilized to verify the 4cm results. Was the surface tested to real survey values or is the 4cm a to the model ties? Are your results horizontal and vertical?
When I test a topographic surface (DTM, DEM, DSM or contours) I will test the analytical aerotriangulation (AT) with blind points and develop a RMSE for a overall math model test. Once I test the math models, I then grade the surface against surveyed cross sections, profiles and well defined ground points, then test and grade to a mapping standard and generate a surface RMSE.
I like to used a 20 or 40 mega pixel camera with a fixed lens for best results. For mapping and achieving the best accuracy I like to use a 50mm or better to gain a better base to height ratio for 3D feature collection in stereo. I have found that using the Samsung with the 50mm (+/-) fixed lens give me stable results. I had to re-write my entire lens models to account for the crop factor of the smaller sensors.
I'm a fan of RTK gps, but not for airborne gps (AB/GPS). I prefer to post process the gps data to ground with static gps for best results. There are a lot of opinions about AB/GPS and this is my Cool-Aid :)
you have great information, thanks for sharing!!!
great info!!!! thanks