Hello,
I have been looking into this ground based photogrammetry in the last couple of weeks. I played with a demo with some of pictures that I took; a basic non-calibrated 14 megapixels point & shoot 200 dollars camera.
I walked along a road, took 20 pictures of a vertical face, pictures overlapping at 60% and then shot easily identifiable control points. Imported pictures in the software, processed the images, aligned the data to the coordinate system with a 3 point solution (total station). Then exported the "cloud" (the buzz word!) to CAD.
Here's an interesting link below to another example from an experienced user. In this case, the camera is mounted on a low cost quadcopter. I was surprised to see how high these things fly! No doubt the FAA is looking into putting regulations for these things.
Anyway, keep your eyes open, cameras are getting better and cheaper. I can some benefits to this technology, that's for sure. It would take a bit to get the most out of it, build good QC procedures (and need a powerful computer) but wow, surprising.
Sounds like the Trimble v10 might be something you'd be interested in.
I have a friend that is a photogrammetrist and has been playing around with ground-based approaches for the past several years. He even made his own UAV before there was any hype around them and proved them out on a few sites.
He has told me numerous times that in no uncertain terms products like Google Tango are just the beginning. Ground-based photogrammetry is very doable, and is on the verge of being very applicable.
As you suggested advances in sensor and computer technology are really all that is currently needed to make a very robust and automatic system affordable. It all revolves around the ability to process a few gigs of photo data and perform the auto-correlation.
I really like the Trimble V10, but I wonder if they'll even recoup their R&D costs before being out-paced by consumer-driven technology.
Quadcopters are nice for surveillance and hovering. They are really slow
for photogrammetry. This Agisoft software is expensive, but the scaled-down
version is cheap but is really set up for one photo.
The standard survey equipment manufacturers do not have their own plane, yet.
So you have the survey equipment manufacturers and their dealers as middle men
jacking up the overall cost.
When the Phantom crashes, you are out $1200 and pay another $1200 to buy
another Phantom. Most of the DJI products are limited by jellying- blurring images
due to shaking of the UAV in the wind.
BTW, Sam Inc and Loomis in Austin have been granted COA certificates.
There's also a need for camera calibration. Trimble has solved that problem by doing the calibration themselves, thus significantly increasing the reliability of the system's accuracy.
You can do it yourself with the appropriate photogrammetric software, but with off-the-shelf cameras, it requires multiple calibrations at different distances. If your camera has a zoom lens, it's still possible but the calibrations become even more involved, albeit possible.
Note that Trimble has DECREASED the possible number of variables in their solution. A general case is significantly more complex, so be careful of what you "assume."
The auto-correlation is dependent on color saturation and color balance. Again, Trimble has solved that variable by "fixing" it with their factory-furnished hardware.
All of these things are possible with the current power of personal computers, but note that having a fixed hardware/software configuration that Trimble has to offer makes it look really easy. It's not as easy with a do-it-yourself system cobbled together by a non-photogrammetrist.
Possible?
You betcha.
Many years ago a local aerial mapping company completed a ground based project and showed the results to our professional association. They were called on to make a 3D study of a historic stone house near a proposed quarry/open mine site.
Mining company was required by law and insurance company to complete preblasting inspection of all occupied structures within a specified distance of the mine site. Owner of the house would not allow any access to the property. Surveyor involved ran control traverse adjacent to the property lines and direct levels to the points. Also triangulated to control points on the structure and computed 3D locations. This was long before prisimless total stations were available.
Aerial mapping firm had a camera designed for ground based photogrammetry that was tripod mounted. They set up on control, established HI of camera and took overlapping photos. The resulting mapping/elevation views and detail were amazing.
With the technology available today a surveyor should be able to gather all the data required without a scanner. However, scanning might still offer a better solution.
Scanning data is far more data rich than photogrammetric data in that scanners offer azimuth, elevation, distance, and reflectance. On the other hand, it's presently slower than imaging a photo. What scanners can do with a single instrument station, photogrammetry needs a minimum of two camera stations and realistically requires at least three. Photogrammetry offers a higher potential accuracy because scanners have a diverging spot size based on distance, and multiple images computed with photogrammetric software offer greater accuracy. Multiple scans still have a fixed spot size and a fixed precision unlike photogrammetry.
Neither is a panacea; it's the professional that needs to choose the right tool for the job at hand.
I agree completey. For the V10 the magic also lies in that the omega phi and kappa of each image from the tilt sensor is known.
In my scenario what enables rapid usability is the development of very high MP fixed sensors, Like the Nokia Lumia. The need for zoom is removed or diminished and only one calibration is needed.
I'm not talking so much cobbled together, instead considering software advances from things like Agisoft. If we are seeing a 41MP sensor in a cell phone, just think of what we'll see in one or two years.
Knowing the omega-phi-kappa is not that important if the initial conditions of orientation are handled with quaternions rather than with a 3x3 direction cosine matrix. The iterative corrections to a standard classical orientation matrix assume that the corrections are infinitesimal. Not so with quaternions, and that can be completely avoided.
The tilt sensors were apparently put in by someone that did not understand that fact, and was going with the fully classical math model. Alan Pope of the USC&GS offered that as an "Alternative Parameterization ..." in a paper over 30 years ago.
> What scanners can do with a single instrument station, photogrammetry needs a minimum of two camera stations and realistically requires at least three.
I really appreciate your insights; i'm thinking a lot about the two (scanning vs photogrammetry) lately.
When multiple scanner setups are required, what is the rule-of-thumb for the required camera setups, as a comparable alternative?
Assuming
S = scanner setups and
C = camera setups,
is it roughly
C = S + 2
or
C = S * 3
or something in between?
> Knowing the omega-phi-kappa is not that important if the initial conditions of orientation are handled with quaternions rather than with a 3x3 direction cosine matrix. The iterative corrections to a standard classical orientation matrix assume that the corrections are infinitesimal. Not so with quaternions, and that can be completely avoided.
This is one of those areas in which I'm content to let other people figure things out, since pretty much everything in Cliff's quote flew over my head. Once the smart people get the concepts into a convenient and affordable package, I'll start looking at using it, making sure to test the results against known positions under circumstances typical to my situation. Just like I do with GPS and EDMs.
:good:
Like I said previously, my friend is the photogrammetrist. I yield on this topic!
You had me at quinine solution, I thought that was medicine (or was that quarter of an onion?).
😀
Two images of a point gives an XYZ coordinate. Three images of a point gives an XYZ coordinate with a check. Four images can make us feel warm and fuzzy. With regard to scanners, they have significant systematic distortions, too. How many scans it takes to get a point cloud that gives you a warm and fuzzy feeling depends on the quality of the scanner's calibration and your level of experience. Same goes for photogrammetry.
> > What scanners can do with a single instrument station, photogrammetry needs a minimum of two camera stations and realistically requires at least three.
>
> I really appreciate your insights; i'm thinking a lot about the two (scanning vs photogrammetry) lately.
>
> When multiple scanner setups are required, what is the rule-of-thumb for the required camera setups, as a comparable alternative?
>
> Assuming
> S = scanner setups and
> C = camera setups,
> is it roughly
> C = S + 2
> or
> C = S * 3
> or something in between?
I believe Cliff will agree the answer comes back to the Fast - Quality - Cost balance surveyors always face. The old quote is you can have any two of the three. If you want it fast and high quality the product will likely be expensive. As with all our tools various models of scanners and cameras have different capabilities and costs. The same is true of software and data processing capabilities.
The tools that are great for one project may require extensive additional work or be unsatisfactory on another project. For example combined photogrammetry and scanning tools sufficient for a major earthwork project may not be usable for architectural study of an ornate historic building.
> This is one of those areas in which I'm content to let other people figure things out, since pretty much everything in Cliff's quote flew over my head. Once the smart people get the concepts into a convenient and affordable package, I'll start looking at using it, making sure to test the results against known positions under circumstances typical to my situation. Just like I do with GPS and EDMs.
Exactly my thought too. A lot of GPS/GNSS users have never heard about single difference, double difference, triple difference. In 2000, a Leica 500 GPS system came with 5 digits serial numbers. Today, the manufacturer use seven digits for the Viva's serial numbers. Put this in your matrix 😉
I understand your background and life as a scholar, Cliff. It is great and I am thankful as technologies require men and women with your expertise level to grow. However, let's not be fooled, more and more folks will get in photogrammetry, as with any other technology... To quote the planet of apes "more will come"...
Depending of the characteristics of the features surveyed, photogrammetry has it challenges: for example, a large industrial diesel tank, dark with little texture could be a challenge without targets. A scanner would probably do a better job.
But a rock face with 1000's of reflective faces and rough texture on a sunny day? A client would probably appreciate the camera-gathered data and lower invoice of the photogrammetry method.
In regards of the camera station, it is simply the location of the camera when the shutter goes click. Might require a bit of planning & thinking, but not really a time consuming task.
The biggest issue with this technology I think is to understand the guiding principles of photogrammetry and photography and apply them. Theory and practice, that doesn't change. Good news, no expensive film to buy anymore, take as many pictures as you want, same price.
Professor Mugnier
what would be an appropriate route to expand my knowledge of close range ground based photogrammetry?
could you recommend some reading, pdf's or books?
how about software? are they generally prohibitively expensive? i am currently looking at photomodeler but academic packages were also recommended: orpheus, solid fit, facade, tiphon, and arpenteur.
as of now, my expertise is in other subsets of surveying, but my curiosity of this is growing
Professor Mugnier
Moe,
download a demo of photomodeler scanner from eos and have a play. i've done the same and found i was able to get up to speed fairly quickly. also if you spend time looking through their tutorial videos and knowledge base you'll get up to speed quickly. also i-Witness and agisoft photoscan have demo versions. looking at the documentation and videos on all three sites should give you a pretty good handle on it all.
Professor Mugnier
I used to teach with my own software, "PC Giant," but students today are too intimidated with MS-DOS software. i now teach with iWitness as it's written by Prof. Clive Frasier, one of the top Photogrammetrists in the world today. PhotoModeler is a popular package with laypeople, but I have never seen anything in the refereed professional journals that reference it. Ergo, I prefer iWitness and iWitness Pro. The "Pro" version is useful if you're going to plan and execute your own photogrammetric project. The less-expensive iWitness is less fancy, and is what I use for professional forensic work when I am handed photos taken years prior. Call me some time and we can discuss the topic further. Tuesdays and Thursdays are best at 225-578-4578.
