Sign in to follow this  
R. Koehler

Directional bias for DJI Pro 2 positions

Recommended Posts

Recently I wanted to confirm the positional accuracy of my DJI Mavis Pro 2. I placed the drone over a known NGS survey mark, then hovered at 2 then 10 meters above the mark. I took a total of 300 images to capture the GNSS latitude and longitude values (100 per height, with about 25 in each cardinal direction , N, E, S, and W) . The DJI Go app showed 17 to 20 GNSS satellites were being tracked. The results are shown below which show a clear directional bias in the positions. Has anyone else run into this? Thanks!

Link to plot image.

https://1drv.ms/u/s!AhKoglIEIVS6g5MzgvCkIu7YEfcm4w?e=BKC8G0

 

Share this post


Link to post
Share on other sites

While most know that the location accuracy of these drones is not on a precision scale, I've not seen anyone quantify it with regard to which way the drone is heading.  Actually, the position variance in your sample only diverge a couple of meters (over all) which is pretty good. 

It is interesting that absolutely 0% reported on the mark.  You would think (hope) that occasionally it would happen to be on target.  Imaging if the plot were bullet holes, the safest place to be would be on the bulls-eye, ironically.

They do cluster like you mention.  Knowing that, if the mechanism existed, an operator could run a similar test, and based upon the results input an offset correction in the settings and end up with largely good (or at least better) position data in the image set.  For now, we'll just have to either employ one of the precision gnss/ppk solutions, or GCPs to correct for inaccuracy.

Edited by Dave Pitman

Share this post


Link to post
Share on other sites

Thanks for your thoughts and input. Yes, I also saw the lack of any points being exactly on target. I'm still wondering how that might happen. For directions, I thing having the GNSS antenna on the rear of the drone is part of the answer.

I agree that surveyed GCPs would be the best solution to this issue. As you also mention, a local real-time GNSS corrector system like RTK tailored for drone operations is an option. Either way means a significant increase in equipment and labor costs. I'd be interested to finding a way to include WAAS/SBAS signals to apply a real-time corrector. I'm still looking for that.

It would also be nice to have a way to post-process the drone GNSS data (like most GPS receivers/software can do), though I haven't been able to find anything online to do this. If nothing else, a person might want to know the directional bias and perhaps find a way to account for this error.

But for those that don't go either of these steps and just use the drone's position for aerial mapping/GIS, this is a serious issue to be sure. From my days as a surveyor for NOAA, the rule of thumb is that your positioning framework should be 10x more accurate than the final geo-spatial product. Using this as a guide, a 2-meter uncertainty in the framework would mean the final product accuracy would be only be good to 20 meters. Now we are talking something a bit more troubling.

Edited by R. Koehler
missing thoughts, misplaced wording

Share this post


Link to post
Share on other sites
52 minutes ago, R. Koehler said:

But for those that don't go either of these steps and just use the drone's position for aerial mapping/GIS, this is a serious issue to be sure. From my days as a surveyor for NOAA, the rule of thumb is that your positioning framework should be 10x more accurate than the final geo-spatial product. Using this as a guide, a 2-meter uncertainty in the framework would mean the final product accuracy would be only be good to 20 meters. Now we are talking something a bit more troubling.

Yep, preaching to the choir!  GCPs or a GNSS kit (just an example https://topodrone.org/) is definitely a requirement for global accuracy.  Otherwise, useful for relative accuracy stuff, like volumetrics, or general reference orthoimagery.

Share this post


Link to post
Share on other sites

There are a lot of variables that need to be accounted for to have an accurate survey or build an accurate 3D model.  

For example: GPS is +/-  16 feet, so if your flying at 7 meters per second, taking into account the latency of the electronics, wind/ground speed, changes in altitude, etc., the drone may have traveled approximately 40ft from where it thinks it took the picture and where it actually took the picture, we call that the zone of confusion.  If the variables were constant then you could calculate the offset for better accuracy.  As you’ve discovered with your own research on directional bias, it’s different between every waypoint. 

Depending on the RTK System you can reduce that zone of confusion to about 20 or even 10ft because they don’t control the drone or they don’t have their own onboard IMU so they can’t account for all the variables. Many of the Drones, like the DJI M210 take the “calibrated” data and use it to navigate the drone more accurately.  No one cares about the drone, they care about the relationship between the sensor and the object being scanned.  

We overclock the GPS, we collect the camera sync from the hotshoe so we know exactly when the second curtain rises (global shutter) or when the full image is recorded (CMOS), we take the telemetry from the IMU, and use statistical error correction to reduce the zone of confusion to about 2ft.  We can process this onboard in realtime at 200-2000Hz depending on the speed of flight and accuracy desired. 

We can calculate more accurately where the drone will be much better than DJI can calculate where it’s been.  We’ve basically turned this into a poor-mans INS, so the challenge is drift.  To counter this we capture the raw GPS data with the image, the corrected GPS data we calculated, and although we can use less we capture ground control points. We can input all of that data into ContextCapture and we’re working on some proprietary tools to pre process drift before using it in any photogrammetry software.  

Do you know Todd Jacobs?

Share this post


Link to post
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

Sign in to follow this