Elevation Problem?

I've been flying inventory flights for a client over the past summer with good results. As part of each flight I fly and calculate volumes for an area that is not disturbed as a control. Over 11 consecutive flights the volumes for the control came in fairly close, generally within 1-2% and at most 4% out. This level of accuracy is fine for what we are doing with the data. In mid October the value for the control jumped up by about 5% and at the end of October by 21% which is unacceptable and obviously raises questions about the rest of the data collected. As far as I can see nothing has changed with the way I am carrying out the flights: same drone/camera, height, overlap etc. I thought maybe there was a GPS issue so I corrected the last flight flight to one taken in August by creating GCP's from features on the ground in the August flight and applied them to the last flights photos. After processing the geotiffs line up almost exactly but the volumes I got were even farther out. I am at a loss what might be causing this, if nothing has changed in how the images are processed I must have an equipment issue of some kind, could the barometer be giving faulty or inconsistent readings?

Dlbeck11 4 November, 2017 23:14

Komentar resmi

Adding GCPs does not affect the shape of the processed elevation model and since making the volume measurement is a relative measurement they don't matter and shouldn't be used.

Our GCP system was designed for non-geotagged images as a way to georeference the outputs. If your camera already has tagged images you should not be using the manual GCP system.

If the barometric altimeter was having issues it is likely your aircraft would not be flying very well since that is the main altitude sensor.

There have been changes to the way the data is being processed on our side.

In general, if you really care about your 3D model quality or elevation data accuracy it is important to use plenty (80%+) of overlap in order to make sure the system has enough data to process things accurately.

Zane 6 November, 2017 17:13

Harap masuk untuk memberikan komentar.

7 komentar

Hi Zane, I've been using 80% front overlap and 70% side. From everything I've read this seems to be fine for elevation modeling

It is really important that I understand what changes were made to your data processing procedures, when they were made and what impact they might have on the elevation model - I need to determine if this might be causing the differences I saw in October. Is it possible for you to elaborate on this?

I remain confused about the use of GCP's and that volumes are relative measurements. Wouldn't using the geotags only result in accuracy as good as the GPS in the drone which could be out several meters? Wouldn't GCP's bring the model back spatially to the real world and allow volume measurements to be closer to absolute values?

Dlbeck11 6 November, 2017 17:59 0 suara

If you care about your results I would recommend using 80/80 as we suggest in Map Pilot and our Data Collection guidelines. Think of what you are doing as 75/75...

We didn't make any changes to the elevation processing.

Volume measurement is a relative measurement. The polygon defines a boundary and all of the volume above the reference surface defined by that polygon gets summed up relative to the surface height.

Zane 6 November, 2017 22:36 0 suara

If I didn't care about the results I guess I wouldn't be asking these questions. No problem going to 80/80 but that doesn't explain why after 12 measurements using 80/70 the volumes suddenly jump by 20%? Doesn't seem to be an overlap issue to me. I flew the same area today with 2 different drones. If the results are the same it will rule out a hardware issue. My only comment on the GCP issue is that if the area of the polygon is off because the gps is out the volumes will be be affected correspondingly.

Dlbeck11 7 November, 2017 05:44 0 suara

Not having enough overlap is a very good explanation for why the elevation products would not be consistent. Maybe you got lucky before with less overlap but for repeatable and consistent results we ALWAYS recommend using at least 80/80.

The other variable here could be how you are drawing your measurement polygon. Make sure you are are using lots of points. Refer back to how you have it drawn in your previous ones and make sure the point placements are similar. Using just a few points to define your polygon will affect how the reference surface is determined and will affect the measurement.

Zane 7 November, 2017 18:23 0 suara

Well if I got lucky 12 times in a row I guess I should have headed to Vegas instead! I flew a trial area with 2 different drones to rule out a hardware problem (there isn't). I then submitted the same photos taken in October (with the 80/70 overlap) to Drone Mapper and lo and behold the results are back to within 1% of what I was seeing before. This tells me you made some change to your system in October that is affecting how the point cloud and DEM is being produced - there is no other way to explain this.

Dlbeck11 9 November, 2017 03:48 0 suara

No changes have been made to our elevation processing or volume calculation process in the last few months. If you want to be scientific about it you should upload the exact same image set that worked previously to see if it comes out the same. All image sets are different. Shadows, gaps, and the way the images line up with the terrain have a big impact in how things get processed. That is why having lots of overlap is required for repeatable results.

Also, are you certain that you are drawing the measurement polygons exactly as you did before? If you would like us to check it please make the maps public and share them so we can compare your measurement polygons.

Zane 9 November, 2017 16:59 0 suara