Bandwagon's Drones Thread

Picasso3

Silver Baronet of the Realm
11,333
5,322
Lol, I wish I had seen this yesterday. I don't know if THIS is why you flew into a tree, but one of my guys did this with someone else's drones on a structure scan.


You need to edit the failsafe behavior first, depending on the density and size of the building. Otherwise, you lose contact when flying around the building, and the drone tries to fly back to the home position.


Exactly what happened, except it hit 10% battery while hovering about 5 feet from me inches off the ground, went into return to home, which was a tree.
 

Bandwagon

Kolohe
<Silver Donator>
23,021
60,316
Yea, that sucks. Just one of those things you have to learn while learning how to use these things. I have a handful of stories like this as well, but they were all on my first fixed wing drone I built. I hit the same fucking powerline 3 times before I figured out what the problem was - the flight mode I was using only uses throttle for altitude control, not elevator.

edit - meant to post my tip as well.
ALWAYS keep one finger on the "stop doing what you're doing" button. Always. Always. Always.

For us & the Solo, we usually hold the controller with one hand while we're just waiting around during a mission, but our finger is still ALWAYS on the "FLY" button.
The last time I crashed, I broke my own rule and hit a tree while doing a real estate video. I wouldn't have crashed if I had left my finger on that button.
 

Picasso3

Silver Baronet of the Realm
11,333
5,322
I'm impressed at how durable the thing was. The arm is cracked but it's flying fine. A10 warthog of drones. Messing around in photoscan now. Does my shit look right?

xGVwhpW.png
 

Bandwagon

Kolohe
<Silver Donator>
23,021
60,316
I haven't used photoscan in almost two years and have way less experience with it than Pix4d....but looks good to me? 47/47 aligned is a good sign. Something you need to get in the habit of doing is looking at the side view of the point cloud after you generate it. There's a lot of jobs that look decent from the top (the orthomosaic looks good), but the DSM is massively fucked up. Spending about 5 minutes searching through the point cloud is a good way to find and address this.

The point cloud SHOULD look like a fuzzy, wooly blanket draped over something. When you start seeing points generated way outside of those bounds or misaligned blocks, you have a problem to solve. The first is usually dirty/low quality lens or poor camera calibration and the latter is usually a geotag or an overlap problem.

Also look at the plan that the images are on. You'll see a lot of variance in the angle of the camera during each picture, but all of them should be level on the same plane unless you're using terrain following. If there is a problem (usually camera calibration), you'll see them look like they're bowing in or out, like a bowl.
 
Last edited:

Picasso3

Silver Baronet of the Realm
11,333
5,322
it's fucking amazing what they can do with technology these days.

Here are the readings I get when comparing this surface to 6 inch lidar which should be really good shit:

AVG_Z_DELTA: -0.901 m
MAX_Z_DELTA: 0.309 m
MIN_Z_DELTA: -1.868 m

I'm wondering how good it'll be if I just adjust the photogram dem down the AVG Z delta. I'm guessing all the elevation info is coming from the onboard GPS exif data? I'm pretty impressed.

bq0M35j.png
 

Bandwagon

Kolohe
<Silver Donator>
23,021
60,316
Mind explaining your "...down the avg delta" part? Pretend like you're talking to someone without a background in this work, because you are. ;)


Yes, it all comes from the drone gps.
I haven't done a large amount of these tests, but try doing a bigger flight next time and seeing the accuracy. In the few tests I've done, the larger sets are more accurate, I assume because the low accuracy GPS data on images is averaged out on larger sets and produces better results after processing.
 

Picasso3

Silver Baronet of the Realm
11,333
5,322
AVG Z delta is the difference between the lidar ground points and my generated tin, i used a boundary within the field area so the trees didn't fuck it up. It's an easy operation to raise/lower an entire tin so if I adjust the whole surface to match the lidar which is +/- 6 inch accuracy, it should bring my stuff more in line.

It'll be curious to see if it's usually off a similar amount so that it can be tuned up without lidar to base a correction off of. (i'm sure the photogram adds depth of grass, which was prob ~6-9 inches high in this area).

As far as large samples so I'll try to think of somewhere, being WV you usually can't go too far without hitting a tree or a building.
 

Bandwagon

Kolohe
<Silver Donator>
23,021
60,316
Ahhhh, gotcha, that makes sense. So the relative accuracy within the dataset is typically pretty good (relying on photogrammetry), but the whole set is being shifted on the Z axis to correct the GPS inaccuracy.


Is there any way I could get you to show me how to do something in qGIS? I basically just want to click around on an ortho anywhere I see bare earth, and have each mouse click generate a point in a vector layer.
I'm still working on automating something like this, but I'm getting some help from a guy when he gets free time.
Until then, I'm just trying to play around with it.
 

Picasso3

Silver Baronet of the Realm
11,333
5,322
No experience with qgis. In civil3d you could click points and have them snap to a surface elevation.
 

Bandwagon

Kolohe
<Silver Donator>
23,021
60,316
I wish I had access to that stuff right now. Oh well, shouldn't be too long. Finishing up negotiations with the engineering company right now, and I'm guessing i'll be putting in my notice at the fire department within 30 days.

Then I get to play with all their software too. ;)
 

Picasso3

Silver Baronet of the Realm
11,333
5,322
I wish I had access to that stuff right now. Oh well, shouldn't be too long. Finishing up negotiations with the engineering company right now, and I'm guessing i'll be putting in my notice at the fire department within 30 days.

Then I get to play with all their software too. ;)

Do you have posts in here regarding the faa exclusion and becoming a for profit business process?
 

Bandwagon

Kolohe
<Silver Donator>
23,021
60,316
I do, but I don't remember what page. My exemption was approved in December of last year.


You don't need to worry about that though. The rules change in 7 days. Look up part 107.
 

Bandwagon

Kolohe
<Silver Donator>
23,021
60,316
You can give me a call if you have some questions.

It's a really good time right now to get involved. You already have the hard part of the required skill set (gis).

The part that used to be really hard is changing in 7 days. I'll pm my number.
 

Picasso3

Silver Baronet of the Realm
11,333
5,322
Damn did you buy pix4d?

I may give you a call soon, i think i have so many questions now i don't know what to ask about first.

I'm hoping my uncle will go to the faa school and knock that out, the nearest one is about 2 hours away.
 

Bandwagon

Kolohe
<Silver Donator>
23,021
60,316
Yea. DroneDeploy is probably your best alternative right now I guess, though I don't know too much about how it's changed in the last year.

2 hours away? Did you check the FAA's list of testing centers? It doesn't have to be at an FAA school.
 

Bandwagon

Kolohe
<Silver Donator>
23,021
60,316
Yea, but there is another list of testing locations for part 107. I can't find it right now though, but I printed off the pages for Oregon and Washington. It's not just FAA testing centers that are doing it.


And I forgot you have photoscan, so fuck drone deploy.
 

Picasso3

Silver Baronet of the Realm
11,333
5,322
I'm the lump in the driveway. This is medium quality, i think i'll do high and see what it does overnight.

nK19PDE.png
 
  • 1Like
Reactions: 1 user