Hi all,

this is a short follow-up on my previous tests (I, II) with u-blox different u-blox M8 GPS modules.
This time I'll report about a survey campaign of several days with different copters and GPS. After processing the data I was surprised about the positional accuracy so I decided to share some numbers.

We used two copters running AC 3.3. One with an M8N and the other with an M8T GPS. The survey was conducted over a period of 4 days.

Six orthomosaics were overlapping:

Dataset   Day time
A         3   10:40
B         1   11:20
C         4   12:00
D         4   10:20
E         4   11:00
F         1   10:00

The orthomosaics were generated using direct georeferencing, i.e. without any RTK processing or ground control points. Hence, the final accuracy, or to be more precise the spatial offset between the maps, is a function of the accuracy of a single GPS and the image processing pipeline.

Since the flights were conducted over 4 days with different GPS and copter setups, the positional offset shows the overall GPS accuracy that can be achieved with M8 GPSs. The processing chain was the same for all datasets. However, there is for sure some influence as well. [I'll provide more details about image processing in my next blog post.]

Flight altitude was 50-60 m. Images were taken using a Canon S110. The size of the survey locations ranges form 5 to 25 ha (12-60 acre). The length of the entire area covered by the orthomosaics (from A to F) is 2.7 km (1.7 miles).

The maximum spatial offset between the orthomosaics is 1.2 m (3.9 ft). The average is 0.87 m (2.9 ft).

Overlay   Offset
A vs B    0.8 m / 2.6 ft
A vs C    0.6 m / 2.0 ft
B vs C    1.2 m / 3.9 ft
B vs D    0.4 m / 1.3 ft
D vs E    1.1 m / 3.6 ft
E vs F    1.1 m / 3.6 ft

Based on my previous tests I expected a much lower accuracy of about 2-4m maximum spatial offset. But it seems that both, the M8 GPS as well as the image processing pipeline, are pretty accurate and stable.

My next tests will be on RTK post-processing as well as the (promised) bench comparison between different M8 versions and antennas.

Best regards,
Thorsten

Views: 3315


MR60
Comment by Hugues on January 13, 2016 at 7:53am

Thx for posting these results. I suppose you did not do a comparison with the standard 3DR GPS+compass module (LEA6 or NEO7) ? I have read some conflicting reports about the fact the M8 is only giving an illusion of beter precision versus NEO7 or LEA6, and I would like to see trustful test data for an objective comparison.

Comment by Carlos Perez on January 13, 2016 at 9:38am

This research is awesome! keep up the great work

Comment by Carl Schaefer on January 13, 2016 at 11:03am

I'm really interested to hear about your RTK setup. Specifically, if you're able to get the Pixhawk to record RAWX messages from the M8T for later use in post-processing and if the M8T is the lone GPS on board, supplying positional information to the pixhawk while also recording carrier phase data, thus eliminating the need for a dual GPS setup.

Nice work as usual!

Comment by Tony K on January 13, 2016 at 11:39am

@Huges

Highly doubt that when designing the M8 series they had "Illusion" in mind. It comes down to filtering of the 1.57542 GHz signal and antenna design. A more sensitive unit will be more prone to interference and not "less accurate" necessarily but more prone to interference because it might not have been designed for DIY drone use. 

Comment by John Bond on January 13, 2016 at 1:07pm

Assuming your methods are such that the "mismatch" is mostly GPS error the numbers you got are to be expected.

A GPS receiver in motion tends to be a bit more accurate than one at a fixed location.  This is counter intuitive
for most people, but it is easily demonstrable by comparing a carrier phase processed track with the code based
solution via common time tags.  This does assume you are operating in a very good satellite visibility area.

Another thing that helps is that your methods (probably) average the GPS error somewhat.

Also, if you did many more similar tests the error numbers would likely increase some, but meter level direct
georeferencing with a single C/A code based receiver is certainly possible.

It will be interesting to hear about your exact methods.


T3
Comment by Thorsten on January 13, 2016 at 1:14pm

@Hugues,

I use M8 GPS on all our copters. From all tests I have done so far the M8 has a higher accuracy. It gets more satellites (systems), get faster fixes and is simply more accurate. And the results posted here are pretty convincing (to me).

But as @Tony mentions, it is more sensitive to interference. I had severe problems with bluetooth telemetry on one setup for example. And there are other problems as well:

https://groups.google.com/forum/#!topic/drones-discuss/4WelUGkEhho

https://forum.u-blox.com/index.php?qa=826&qa_1=timing-issues-m8...

So you have to carefully test your setup. But I guess this holds true for all setups sind I assume that no GPS was designed with drones in mind ;-)

Anyway, it would be good to compare it again. We are not too far away from each other so maybe we could work on this together.

@Carlos, thanks!

@Carl,

Yes it is working! I hope to post some report in the next two month or so with similar data as shown here.

Comment by earthpatrol on January 13, 2016 at 1:19pm

Nice work. Do you take into account weather related variances as well? Wind, PV=nRT type considerations? Or is it  a wash/normalized based on the sampling methodology? Keep up the good work and thanks for sharing.


T3
Comment by Thorsten on January 13, 2016 at 1:20pm

@John,

I was thinking about the averaging effect during photogrammetry as well. 

However, since the flight are from two different setups and over a period of 4 days I think this it not the case. At least it would be kind of magic.

The trick I guess is the precise georeferencing of the images below 100ms. In this case it seems to boil down to GPS accuracy. What is most interesting I guess is that there is no larger offset over they days. 


T3
Comment by Thorsten on January 13, 2016 at 1:26pm

@earthpatrol,

thanks!

No, no PV=nRT or something :-) The data is based on single flights which I processed fully independently. There is no relation no overall equation, adjustment or whatsoever.

We even did not let the GPSs warm up (which once resulted in a short SBAS "flyaway"...). 

Comment by Darius Jack on January 13, 2016 at 4:31pm

Could you tell me if real-time RTG GPS is supported by Mavlink for data upload or should I use another radio ?

Comment

You need to be a member of DIY Drones to add comments!

Join DIY Drones

© 2019   Created by Chris Anderson.   Powered by

Badges  |  Report an Issue  |  Terms of Service