I am wondering if anybody has been able to do any sort of characterization of the attitude estimation accuracy of the APM 2.0. I asked on the Blog Post 'Great lecture on the Outback Challenge by APM developer ... what type of direct geo-referencing accuracy he was getting and he said "20m error at 100m altitude in our testing so far." If one were to assume all the error was from angular uncertainty in a single axis, this would suggest a number in the realm of 11 degrees of error. This is one case, and I know there are a lot of contributing factors for direct geo-referencing error but these are only numbers I could find so far. Does anybody else have insight on what one could expect? Are there any special calibration or alignment steps taken at the 3DR factory?

Views: 2163

Comment by Andrew Tridgell on April 29, 2012 at 9:05pm

Hi David,

Much of the geo-referencing error in our tests is from timing. We need to match attitude from the sensors to the camera frame time in a plane flying at 30m/s and being buffeted by the wind.

I think the attitude error when close to level in APM is probably less that 5 degrees, but it does depend on a lot of factors, and its hard to get a reference vector to check against with that sort of accuracy.

Cheers, Tridge

Comment by David on April 29, 2012 at 9:14pm

Thanks for the additional info Andrew. Yeah it can be pretty tricky getting everything synced up to give a precise position/attitude reading at the moment if capture. Even 5 degrees is pretty rough for doing any sort of direct projection. Do you think the best way to characterize the accuracy would be by flying it in an aircraft that has a high accuracy GPS/IMU and compare the recordings? 


Comment by Andrew Tridgell on April 29, 2012 at 9:52pm

The closest I've come to measuring this is by having two APMs in a plane. In that case the difference between them tends to be small, except in sharp turns.

Here's an example with an APM1 and an APM2 in the same plane:

Comment by David on April 29, 2012 at 9:57pm

That is great graph and a cool experiment! I am impressed that they correlate that well.

Comment by James on April 30, 2012 at 6:28am

Hi David, nice post!

I've been doing some calcs for auto georeference using the autodesk method (rotation and scale vs gis extend points). the wicker points in georef photos from uav's, are photos gps center and as you mentioned omega phi kappa time matching, take a look at the best case scenario in the diagram . 

we see that for a 200m height flight, then only error leading parameter is gps accuracy. In the unlike event of flying in a 20degree slope terrain or bankining 20 degrees, the error is about 5-10m as you mentioned (see diagram 2) but if you have great quality DEM or enough overlap to create your own then you can georeference and then orthorectify using DEM: 

Then again, you can fly in a lower altitude (100m) witch is the legitimate for most countries that i know of and the error falls down to 0,5m!! Also you can add a gimbal stabilizer and worry only for the sloppy terrain :

 for the above calculations, i used the specs from a canon ixus 100is/ 780sd with 5.0 focal length (no zoom)


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