Small Details Matter

We have been running RTK for a year now and with a little testing and DroneDeploy’s support we have the algorithms dialed in and beta with Yuneec has their solution in good shape with a few enhancements on the horizon.

DroneDeploy

The main reason for this post is backup of what happens when you fly in the wind but to add to the problem what happens when using oblique images. It was only about a 12-15mph wind but at altitude the gust were probably more around 25-30mph so as you are trying to maintain a consistent attitude you get blown of course. When this happened before our Yuneec H520E RTK you could see the errors visually on the cam relocate report but with the RTK the errors are minimized because the enhanced GPS can react quicker and the coordinate is so much more accurate that processing can reconstruct much more accurately. Here’s the two pieces of info and the icing on the cake which shows off the RTK accuracy.

Processing Report

This project was flown at 100ft w/ 75/65 overlaps and an 75% gimbal pitch. Every map that I have submitted with obliques up to this point I have seen about a 10% increase in error which with the RTK isn’t much on the report but with non-RTK that error goes from being based on 2 inches to about 18 inches or 2.2in and 19.8in RMSE. Even with RTK when you are fighting for every 0.01ft you can get.

One last thing we have noticed which may be causing these effects to be worse on the Yuneec is the fact that the camera is bigger and the Y axis can move 360 so wind can definitely push it around. This effect you will see on the camera distortion portion of the Processing Report. A 5% intrinsic error is NOT good. Should be more like 1% +/- 0.5%.

1 Like

Hello Michael,
thank you very much for that detailed report. So the wind or gusts seem to cause the deviations. Maybe there is a way to reduce the effect by altering or adapting the routing?
As you may now that I waited since Q1 of 2019 for my H520 (in use since Oct 2017) to finally get its “promised” RTK retrofit, but it never came. I carefully read all your reports on “H520E RTK From the Box Up” since April 2021, great work and expertise! But last year in November I decided to switch to another platform for RTK/PPK for RGB and thermal mapping, which I did with a heavy heart. But the H520 remains in operation for other purposes.
The new UAS will arrive next week and I will report here and there about my experience and results within DroneDeploy, which is IMO the most reliable platform especially in terms of “stitching everything better than anybody else”.

1 Like

I think it definitely affects the E90X camera more than our other drones probably due to the larger surface area and the fact that it can be affected in yaw movement since it can rotate almost 360 degrees.

What platform are you moving to? We are still debating mostly dependent on their engineering team’s ability to resolve the list of issues I sent them or the M300 RTK. The M300 RTK is the leader right now because we can do Lidar and RGB at the same time. I didn’t realize the 20MP camera on the L1 is basically the same sensor as the P4P but an electronic shutter. We could always grow into adding the P1 but if we are getting Lidar terrain and depending on how manageable it is the RGB would probably be reduce to just an ortho. It just seems like a comparatively inexpensive way to get into Lidar but have the ability to scale to much more functionality.

1 Like

Hello Michael,
the platform I am moving to is the “Autel EVO2 Enterprise Dual 640T RTK”. This would enable us to offer RGB+IR-mapping in a single mission with astonishingly good RMSE values (and geo referencing) in the first run. Even IR-models for relative temperature depiction will be possible with oblique IR imagery. I am not sure about the results of the RGB camera and if the 48 MP can make it into the point cloud calculation. Probably (as of NOV/DEC 2021) automated missions will only run in 12 MP, but 48 MP can be handflown with a 2 sec interval (Jpegs).
An M300 is beyond my budget but is for sure the most cost effective way to get into LiDAR with an “L1 plus P1” setup. Also LiDAR is beyond my budget.
So for the time being the EVO2 Enterprise - as being placed by Autel Robotics between the Mavic Pro and the M300 (quoting CEO Maxwell Lee on dronedj.com from Jan 12th 2022) - is exactly located in our budget range with a good value for money IMO.
I will report on my first “dual” missions (2D and 3D) using RTK within DD, as soon as I get everything up and running…

2 Likes

There’s no doubt about that. It’s taken us plenty of years to get to the point where we can substantiate more than $20K on one unit. I will tell you that the Evo II RTK series is very good. There are some posts on here with accuracy reports and it’s every bit as good as any other RTK drone that is out there right now. DJI needs a Phantom RTK 2 before I would even look at them.

1 Like

As a daily user of the Rock Robotic R2A I can tell you without a doubt that it outputs better results than the L1. The L1 gave me hobby kit grade LiDAR scans and the R2A by far outperforms it in every way. Orthos look great out of the 24MP a5100, the point cloud has a sub 5cm cross section even at high/fast altitudes, and the Rock Cloud is by far the best way to share (no account needed for viewing), visualize, and manually edit a point cloud.

If you are thinking about an M300 give this some serious thought because although it is more expensive than the L1 it delivers much better data from the same sensor and you can make payments of <$200 a month to get it in your door this week. It’s definitely worth a look!

https://cloud.rockrobotic.com/share/f44c7b2c-97b2-4829-b95e-12cca8795314

2 Likes

Thanks for that! I was actually interested in the R2A so it’s good to hear positive feedback. If I recall it’s around $60k for the package? I lean towards the L1 just for convenience and cost. We do some pretty crazy stuff with photogrammetry point cloud editing and I figured Lidar would still be way better than that so did I really need to spend the extra to go 2 levels up or was the photogrammetry to lidar move enough in of itself.