Manual vs CrossHatch for volume calculations

[BrianBulla]

Hi,

So typically I fly manually and take pictures at all sort of different angles, distances, etc. for calculating volume on gravel and asphalt that we stockpile.  I thought I'd compare the same pile using a crosshatch survey and the results were quite different.

Are there any adjustments to the settings in Site Scan I should use for attaining more accuracy?  Using crosshatch would definitely save time on these larger sites with multiple piles to drone.

Here is what I got via Crosshatch:

And here is what I get flying the drone using a "Point of Interest" pattern in DJI Go 4, combined with manually flying some pattern over it as well.

Over 150 cubic meters difference is signifigant for what is a fairly small pile.

[Pål_Herman_Sund]

Guessing... one or several vertexes are made "on top of surface" which makes the base plan wrong hence too small volume calculated in the crosshatch calculation. An easy source of error especially if the terrain is not flat (or close to flat). Did you try to compare the 2 rasters/DSM's to see if there are systematic differences between them?

[BrianBulla]

Hi.  Looking at the mesh that is generated for both, the Z values at a given location are different, but the difference between what is (more or less) the 'base' of the pile and the peak of the pile are pretty much the same....about a 5.6m difference in both cases.

So I guess the differnce in volume calculations would be more related to the detail in the manual flight as opposed to the hatch flight??  With the manual flight, I took 72 pictures distributed like so....always pointed at the pile with varying camera angles.

And with the hatch flight, the software took 321 photos over the entire site, but how many of those were focused on the pile could take some time to figure out.

[AndrewCarey1]

Hi Brian, I would strongly advise that you perform pre-planned autonomous flights to maintain consistent image capture.  By capturing images using a crosshatch or area survey method the image overlap, altitude, and distance to the piles will produce a more consistent product.  Measurements within consistently processed data sets can be more reliable and compared to future changes at this site.

Additionally, I would strongly advise the use of ground control points, when possible, as integrating these known locations will provide your data sets with additional statistical values for added spatial confidence. 

[Pål_Herman_Sund]

Supporting @AndrewCarey1 comments on consistent image capture, re-use the same flight plan over and over again, and the use of ground control. This pile is not very large and area-survey flightplan with a high overlap (like 80/80) should be adequate. Define a project area and plan for stereo-capture outside the defined area. For further troubleshooting the processing reports could be fine to have a look at.

But - the difference between the volumes here is very big (~ 20%)  I would first try to have a look at the "base plane".

I have seen examples where Site Scan does not automatically choose to calibrate camera focal length which might produce a "(z) shifted model" but here you are looking at relative figures (top pile - base) so I will be surprised if a non-calibrated focal length would be the source here. 

[BrianBulla]

Hi.

The folks at esri Canada who set us up with Site Scan have talked to me, and suggested using GCPs and flying only with the camera pointed straight down (nadir), would be the recommended practice.  I currently have no access to any equipment for GCPs, so that is something we are looking in to.

As for nadir, I have always done my manual flights with some sort of oblique angle....like 30 to 45 degrees.  I've never done it just stright down, so I'll be interested to see what sort of difference that makes.  Even when you do the Cross Hatch in SiteScan, the default is an oblique angle.

Once I get the GCP issue sorted out, I'll try things again.