Agisoft Metashape
Agisoft Metashape => General => Topic started by: sigma on May 01, 2012, 02:26:10 AM
-
I am considering evaluating Photoscan Pro for an architectural project.
I would like to generate a 3D model of a valley using an array of high res aerial 2D photos. The valley is approx 1 km wide by 2 km long. I want the output model to give me a resolution of 5 meters or better so that it can be imported in a CAD application and worked on by architects.
1. Is this possible with Photoscan Pro?
2. how do you recommend I take pictures to achieve that resolution (how many pictures, at what angle, etc)?
3. At what altitude should the photos be taken, ideally?
4. Will a 7MP camera be sufficient for this or should I use a better resolution?
5. Should I use the optical zoom to achieve better results?
6. How can I determine the ideal number of pictures to be taken?
Thanks!
-
Sigma
This needs a proper mapping engineering planning to do this, to calculate the strips, the overlapping between photos, etc.
To do this you should tell which camera do you have available and which lens.
If you want I can help you on this, I am a cartographer engineer with 14 years experience in aerial photogrammetric/GIS projects.
-
Hello sigma!
Maybe you even don?t need the Pro-version because of your wanted size.
1km*2km at 5m resolution will be a 200px*400px (=80000px) heightmap (really small). 1km*2km at 1m resolution should be possible (1000px*2000px = 2000000px) heightmap. And the object could have an 10000px*20000px texturemap which means one pixel for every 10cm?. This should be enough for your project I think. You could generate 3D-objects with 5000000 Polys and much more texture-size - it depends on your computer.
A 7MP-Cam could be enough for this if you shot 100 Photos or more. More Photos means more quality and more MegaPixels (without noise) means more accuracy. For heightmaps it?s good to use the same altitude and same angle looking to the ground. Just try-out PhotoScanStandard with your 7MP-Cam and photograph some old buildings or walls etc. - you will be fascinated by the precision.
A better Cam will give you better results. Take a look at the new SonyAlpha 57 (16MP) or SonyAlpha 65 (24MP with builtin-GPS for use in the PhotoScanPro).
Optical zoom gives you better results (less lens-distortion) if you don?t want to undistort photos for fast creating 3D-Models (maybe of some trees or stones in the valley) . But you?ve not to change the zooming until the session ends.
If I were you I would contact fpbv!
All the best.
Greetings
tezen
-
Thanks so much for the replies!
Indeed, it makes a lot of sense to try it out first.
So here's what I will do: First, today I will take pictures of another geagraphical feature (a smaller landscape) and try to import them in PS. Then, provided the results are encouraging, I will plan my visit to the valley with the help of fpbv (thanks for the offer!), considering the lens, height, etc.
My alternatives at this point are to hire a lidar survey (~ $30,000 USD) or a field visit with a topographer (3 ~4 weeks to take physical measurements with a handheld GPS). So I would much rather make this approach work!
10 cm resolution would be overkill, but if the data is there, I'll certainly use it. 1m should be more than adequate. Do you think that 5,000,000 polygons will be enough to map an area of 2,000,000 pixels, considering that there are large variations in the height of the features on the land (peaks more than 1,000 meters higher than the center valley) ?
-
Let me make you an offer:
we will do the whole work for you for 15000 $ and you will get 2 cm ground resolution.
Oliver
-
Hello Sigma!
If my client would spent me so much dollars (and time) for this project I would do this:
Buying a big quadrocopter (~10000$) for a high-quality remote-controlable-cam & gps-system (like Sony Alpha65 for ~1500$) for taking thousands of photos from different perspectives of each valley-wall and the valley itself (Btw for your try-outs: Cloudy diffuse sunlight at the middle of the day is the best and there shouldn?t be winds for natural photography). With a nice PC (~5000$) there will be good PhotoScan-results over night with more than ten millions of polygons. Which means for roundabout 5km? (2km? ground, 3km? hills and trees) more than two polygons per m?. Thats enough if you think about flat areas (with less polygons) and a higher resolution texture(s) on top of it. You could calculate even more polygons by changing the setting or by splitting the ground, hills and some interesting parts (like rocks) into "chunks". For more releastic results I would built up a low-resolution 3D-background around the valley.
Of course you?ve to retopologize the high-resolution-results for better use in other rendering-software. Afterwards you?ve big equipment for future projects. But it could be hard work.
2cm? ground-resolution means 12500MegaPixel-Texture, 12500 Million Polygons and even more?!
Greetings
tezen
-
Oliver: thanks for the offer, but I think I'll do this myself... :)
Tezen: Indeed, the price is ridiculously high, hence the reason I'm looking for alternatives. But I like your wish list: the quadrocopter sounds like a very cool gadget!
Went out and took pictures of a rock formation today. The feature is approx 500 meters high, and 700 meters wide. I used a cheap 8MP 'pocket' camera at full resolution. Only took 17 pictures, from the road, about 700m away. Sun was setting, so not all pictures are very clear, which somewhat affected the output model.
Everything processed very well, and PS generated almost 2 million polys (ultra high quality, smooth geometry). I need to go through some of the tutorials to see how things can be improved, but so far, I'm very pleased with the potential.
How can I assess the resolution of the generated model? The software tells me the area is 1.2 m^2. I'm guessing there's a way with GCP to correct that, but I won't do this for this test model. But is there a way to determine the height of one polygon with respect to the overall height of the generated model?
Thanks again!
-
Hello Sigma!
Use a good DSLR-cam instead of a 7mp-pocket-camera the results will be much better because of a bigger&better sensor. Take a look at this picture for all those different types of cam-sensors and you know what I mean: http://de.wikipedia.org/w/index.php?title=Datei:Sensorformate.svg&filetimestamp=20081203074952 . Just borrow one DLSR-cam from your friends or family and compare the results.
If you own a good HD-Cam: Just take a walk around an object and cut the recorded video into roundabout 100 (sequenced) pictures. It works - not really good but it works if you want to scan a high number of objects in short time. Take a look at another software (not released yet) which shows this technique: http://www.vi3dim.com/ . For sure PhotoScan will give you better results if you use 600 (!) HD-cam-pix.
Ask Alexey about your scale-question. Don?t know...
Greetings
tezen
-
Hi tezen!
yes, for the actual photo shoot, I will be using a Canon EOS7D with a Canon 18-55 mm lens. That should improve the results significantly.
fpbv,
if your offer still stands, I would very much appreciate hints/tips on how to approach this project as professionally as possible. Although I have a decent camera, the one I have attached to a UAV is of very poor quality.
My initial thought was to take pictures of the periferal mountains from the ground and take pictures of the ground from the mountains on the side. This way I would attempt to get shots that are as close as possible to orthogonal to the surface. I will also try to get 70%+ overlap between shots to make it as easy as possible for the software to stitch the tiles. If that doesn't produce good results, I can use the UAV, but the quality of the photos is barely in t eh2MP range so quality will suffer tremendously.
Anything else you can recommend? Optimal distance from the ground, camera settings, etc?
Thanks!
-
If you own a good HD-Cam: Just take a walk around an object and cut the recorded video into roundabout 100 (sequenced) pictures. It works - not really good but it works if you want to scan a high number of objects in short time. Take a look at another software (not released yet) which shows this technique: http://www.vi3dim.com/ . For sure PhotoScan will give you better results if you use 600 (!) HD-cam-pix
This can work quite well as a way to quickly generate alot of images. I use ffmpeg http://ffmpeg.org (http://ffmpeg.org) and the following command to split into still images:
ffmpeg -i input.avi -r 1 -s 1920x1080 -f image2 frame_%04d.jpg
Thought I'd share as it took me a while to figure it out.
input.avi =: movie file - doesn't need to be avi though
-r 1 = rate i.e. 1 image per second
-s 1920x1080 = dimensions of output images
-
Hi tezen!
yes, for the actual photo shoot, I will be using a Canon EOS7D with a Canon 18-55 mm lens. That should improve the results significantly.
fpbv,
if your offer still stands, I would very much appreciate hints/tips on how to approach this project as professionally as possible. Although I have a decent camera, the one I have attached to a UAV is of very poor quality.
My initial thought was to take pictures of the periferal mountains from the ground and take pictures of the ground from the mountains on the side. This way I would attempt to get shots that are as close as possible to orthogonal to the surface. I will also try to get 70%+ overlap between shots to make it as easy as possible for the software to stitch the tiles. If that doesn't produce good results, I can use the UAV, but the quality of the photos is barely in t eh2MP range so quality will suffer tremendously.
Anything else you can recommend? Optimal distance from the ground, camera settings, etc?
Thanks!
Sigma
This camera that you are using is good enough to to the job.
The main concern is the lens, this zoom lens for aerial has some problems, you should have a fixed lens to do, something with 24 or 35 mm lens.
Regarding ground sample distance - GSD - you should see how much do you need: 1 meter, 20 cm, 5 cm, etc.
When you decide how much this will fit to you so you gonna calculate how the optimal distance from the ground and how many strips.
About the 70% longitudinal overlap is ideal to do this, and in addition to 30% or even 40% lateral overlap between strips, you will have a very very good terrain modeling.
In other way you can fly shooting obliqual imagery but you should take care to no have too much skyline.
I can tell you, both ways you will save money and time. If your UAV could carry a point-shoot camera with 16 megapixel so you could have a very good results to.
Don?t try to use 2 mp, you will suffer a lot, I did this with a celular phone but it didn?t work well.
So, just tell us which plataform you want to use: plane or helicopter?
-
I can try to find a lightweight 16MP camera and attach it to a plane (UAV). It will take time to get the equipment and assemble it and test it, so I would much rather not go that route if I can avoid it.
As described earlier, the site is a valley surrounded with high mountains (think of it as a bowl, more or less). So, I think I will climb up the mountains and take pictures, hundreds of pictures, from the top of the mountains around the site. I will try to capture a little bit more than half of the bottom section which, as I turn around the 'bowl' should give the software some overlap between all tiles, and capture the site with an as close to possible aerial perspective. I will also get a GPS device to record the exact location of each picture.
Then I could repeat with the sequence from the bottom of the bowl, looking up towards to top of the mountains to help with capturing the mountain tops.
Will the software be capable of 'merging' both set of photos/3D models? Any special processing required to do that?
Also, since this cannot happen very quickly, the sun will change position during the photo shoot. Will this have a significant impact on the results?
thanks again for any insight!
-
Best way is to sit in airplane with a good camera, give instructions to pilot WHAT you go shooting.
BUT Oli63 give you a reasonable price because you need a few "tricks" or better be handy with all this and he can teach you a lot of things for your next work like this....
Afther more than year of day-to-day use can say, i know the software good, and it was not my first one..
Personaly if you have budget use use its services and spare few bucks for equipment that you will use later. if done profesionaly the output can shock your clients= more work for you...... that is what im done and its bring me more work....
Sound reasonable ??
Think on that you must climb a lot, take equipment, and later you will find that is not what you wanted.
In what country you live - need to shoot ?
-
Wishgranter said it all.
The best to do this is to use a airplane.
If you want to I did a similiar valley recently, so I can send you a sample.
-
Hello!
I just checked the prices for Quattro- and Hexacopters which are able to carry a 1kg-Cam and for my surprise the prices are between 500$-1000€. Just google for "Gaui copter" or "DJI copter". But depending on the land you?re living in there are a lot of laws and rules. F. E. n Austria it?s relative strictly forbidden to use an air-vehicle with a built-in cam. In my country (germany) you need a special insurance, for commercial use even some licences and so one...
Greetings
tezen
-
Tenyen, the copter for 1.000 are useles, that you mark as you will use it in real enviroment.... but i can be wrong, but im in contact with comunity around it so i know what im write.
For "semiprofessional" use you need at least something in 3.000 euro range and they cannot handle big DSLRs.... use with one of my friend lately....
another thing is to fly with it, need few weeks of testing at least if not months. Battery is another thing they can be used just about 100-150 times and cost something.
im personaly use services of experienced friend with hand glider - sit with him and go shoot what im need, is better than Airplane - lower atlitude fly ( 10-15m :-) ) but every job need another solution.
For walley like mentioned here is best the Airplane, quadcopter is for buildings.
And im trying use services of someone experienced, it cost something but when you work on few project over year is better to use someone services that have experience and equipment....
-
I am amazed how simple people imagine the process of developing good 3D models to be. With the mentioned methods you are lightyears away from ever getting something useable!
1. You definitely do need an autonomously flying system as nobody is able to remote control a UAV from the ground precise enough to make pictures in equal distances from each other along a STRAIGHT line (not to mention many parallel lines, including constant overlapping in both dimensions). So: forget it!
2. Besides that, it is pretty challenging to develop software that is able to control the UAV in such a way that it flies along lines with a horizontal and vertical deviation less than 2 (vertically 5) meters, including compensation for wind and turbulence. If you cannot, there will constantly be holes in the pattern, as well as too much and too little overlapping to get good 3D data. It took us quite a while to handle this problem.
3. You need a microcontroller on board that is able to release the camera shutter depending on the UAVs position. Otherwise wind will have a great influence on the stability of the picture overlapping factor, which again destroys the 3D data quality. The controller needs environmental data like true airspeed, wind component and so on. Besides, the camera has to be able to fire quick enough. The alternative to set the camery in continuous fireing mode, shooting 2-10 pics per second, won't make you happy as well, as you will get a mountain of pictures which certainly will bring your PC and Photoscan down to its knees.
4. Think of the required payload! The UAV has to be able to carry at least 400 g payload. The mentioned toys are miles away from providing it.
5. You need mission planning software, data about the optical and other performance factors of your camera and lens and have to match it with other control factors of the plane and hardware.
6. It is more than helpful to have a bidirectional data link between UAV and ground station.
7. Be sure: the law is you very smallest problem! In Austria it is - like so often here - not really forbidden, but also not completely allowed. Anyway, they are going to adapt it currently.
8. I don't see how a pilot or a second person on a "real" plane could be able to make pictures which obey all these required conditions. You will always have parts of the plane on your pictures, which alone can drive you crazy when putting the pics into PS. Not to talk about how hard or - in fact - impossible it is to fly along straight lines with minimum deviation without some special instruments. I know what I am talking about, I am holding an airline pilots license.
9. - 30. .............
With other words: much better you tell an expert what you need, pay him some money and get a very high quality result. Even more, as you seem to have more money than is necessary. :-)
Greetings
Oliver
-
Hi Oliver,
I totally agree with you, I am an expert in digital cartography and photogrammetry with 15 years experience. It turns out that today, many people believe that generate digital cartography is something light, without any kind of professional training can go with your camera phone and generate cartographic models. In my country I see as unfair competition is presented, where all appeared says it can generate maps with a camera attached to a bird and offers services like a cartography company.
And the saddest thing is that most people do not have any kind of professional training in photogrammetry.
Best Regards,
Diego
-
Think on this: Photoscan is a GREAT tool, when you know what to do, how to prepare data, then output is excelent.
As Oli63 and Diego wrote, is not for playing with it without a lot of knowlege and experience..... If it was so easy then services in this area would cost not so much......
Oli thanx for sharing deeper info..
Sigma as im wrote earlier, you can learn a LOT from profesional like Oli for your career if want grow in this area..
-
Oli63 and Diego
You are all right!
Here in Brazil the game is way unfair too.
Unfortunaly there?s a lot people doing "mapping/cartography" without knowledge.
Myself thought to give up to working in this area but I decide to not do since I discover some new technologies like PhotoScan.
I hope the future will be bright for us professionals because the competition is really not fair.
-
Well, life is unfair. But not always to your disadvantage. ;-)
Isn't being more innovative, quicker, smarter, more effective.... than others the really interesting thing in life?
Give others the chance to do it the wrong way too!
-
Hello!
Sigma wrote that he wants to generate a model of a 1km*2km valley with allmost 5m precision.
I think that and even better solution (1m) could be done.
@Wishgranter:
Yes - professional Quadrocopters are at another pricerange (~10000).
If you once handle a RC-helicopter an RC-Quadrocopter isn?t a big challenge to fly with.
The battery-cost is low if you compare it to helium/baloongas used on balloon/blimbs.
Take a look at the DJI Hexacopter Flamewheel 550 on the net and videoportals.
I`ll use it mainly for shooting Videos - for sure testing it with PhotoScan.
@Oli63:
1. There are some solutions using GPS.
2. Point 1...
3. A remote-controlled Cam will work. In continuous fireing mode you could select the best shots.
4. The DJI Flamewheel 550 (Hexacopter) could take up to 1kg.
5. You?ve to do the same things even on the ground.
6. Yes.
7. I`ve heard that there could be much problems in germany (except bavaria).
8. Point 1...
9. - 30. ... :D
Experts/professionals with lots of experience in cartography and photogrammetry will do the job better for sure. This summer I?ll show my results on this forum and then everybody will see if I was going the wrong way!
@diegotorres:
Don?t know to who you?re talking to (Sigma or me?). Your joke is OK if it?s sick attaching a Remote-controlled DSLR (16Mp) via gimbal to a Hexacopter. Btw: I`ll use the results for artistic purposes and not for exact cartography with the precision of two inches in an area roundabout 1km*2km...
Greetings
tezen
-
well, I took some pictures of a section of the valley over the weekend, and the results are good but not great. I would tend to agree that aerial photography would improve things quite a bit. So I will get a plane/copter to do a flyover with a good camera.
You are inferring that a plane might be more appropriate that a copter. Am I reading this right?
I also noticed that tagging each photo with coordinates improves things quite a bit, so I am on the market for a good UAV with GPS and ability to mount a good camera on it. I'll be going through forums to get recommendations for such a setup, but I'm also open to suggestions: please let me know what you are having success with along with lessons learnt.
Thanks again!
-
see PM......
-
Hi sigma,
I am getting good results - much better than 5 m resolution - using a canon D10 ($250) bolted to the flightstep of a Cessna 172 with $50 worth of ram mounts (B-231ZU base, short arm, 1" ball).
I used CHDK to boot a modified firmware for the camera and run a 3 second intervalometer script I wrote, with shutter speed of 1/1500s and the neutral density filter forced off.
Using Photoscan Pro I am able to generate good quality DSMs and orthos, just by setting the ground plane with coordinates georeferenced from best available data that I can ID on the photos. I am using some LiDAR to get the z values, and 6" aerial imagery from bing maps to get the xy. The DSMs need some cleaning (I have lots of noise where the water is), but that might be better done in CAD or using meshlab, or something like that.
-
Have a close look at
http://diydrones.com/forum/categories/aerial-photography/listForCategory