Relative accuracy with drone images withouth GCPs

[AD27]

Hi All!

Currently my job is to measure rooftops with a maximum error of 10 cms. I have a DJI phantom 4 Pro V2 drone and I flew a reference building to see if I can produce accurate measurements. I used Pix4D capture with 3D model mission with 45 meters height and around 2 cm GSD and  80% front and side overlap and I had around 270 pictures. After importing and going through the aligning process and point cloud, mesh, textures and ortho photos, I had around 2-3 meters error in the measurements compared to the reference. What can I do, what did I do wrong? If you need more details, please ask for it.

Thank you for your help in advance,
Adam

[dpitman]

Hi Adam,

If you mean that a linear measurement in model space was 2-3 meters different that actual conditions, that seems high.  What kind of lengths are we talking about?

Are all of the photos oriented nadir?

[AD27]

Hi dpitman!

Yes, a linear measurement on the 3D model and orthomosaic is off by at least a meter, I had reference measurements 41,7 meters long as well as an 8 meters one as well.
Sorry, I'm a newbie in this software as well as the measurements, but what does nadir mean?

Thank you,
Adam

[dpitman]

Nadir means that the camera was pointing straight down with no oblique angle.

Using a P4P with no GCPs, relative linear accuracy is generally within 3%.  So, if you have an actual length of 41m and the model shows a 1m difference, that is less than 3% error which is not large.  To try to tighten it up, in MS you could create Scale Bars using the corners of the roof planes, or other geometry in the scene, for which you know the exact values.

Currently my job is to measure rooftops with a maximum error of 10 cms

Realistically, I think you would need to state your tolerance as a % rather than a fixed value.

[AD27]

The camera was around in an 80 degrees angle, so not in nadir, could it be the problem? There was also 1-2 meters difference with a shorter 8 meter parts as well. So, if I want to achieve maximum 10 cm error I would need GCPs or an RTK drone?
How can I tighten it up exactly with scale bars?

[Paulo]

Hello Adam,

2 m on a known 8 m distance seems very high. Maybe you can provide more details on your alignment result. Either the pdf report or the Tie points info from Workspace cntext menu. Maybe also look at the reference pane and review camera GPS  errors....screen capture can help.
As Dave said if you introduce scale bars into your project (minimum 1. better 2) then you should get a correctly scaled model and measured lengths on rooftop should be correct to within a few cm. But even without scale bars you should not  get a 2 m error on a 8 m distance. . To define a scale bar just place 2 markers at extremity of a known distance, select them and use context menu Create scale bar ... then enter known distance....Update transformation...see https://agisoft.freshdesk.com/support/solutions/articles/31000148855-coded-targets-and-scale-bars#How-to-Create-Scale-Bar-and-Set-Up-Reference-Distance

[AD27]

Hi Paul,

Thank you for the help, please see attached the pdf and screen capture, hope this helps a bit.

Thank you,
Adam

[Paulo]

Hello Adam,

the report shows quite a good alignment. However, I would do camera optimization (star button from reference pane), unchecking b1, b2 as they show extremely high correlation with f, cx and cy. After optimization, recheck yur known distances...