Help! How can I increase accuracy of DEM without GCP?

[tabermiyauchi]

Hello,

I am creating a DEM of a very small surface on the order of millimeters using a micro lens. I am using a multi-axis traverse to move the camera above the surface to capture only vertical pictures. I do not have the camera coordinate positions. The dense point cloud turns out very clean and I am able to scale the model correctly in the X and Y direction (refer to image), using a horizontal line between points 1 and 2. However in analyzing the model, I have found that the model is not scaled correctly in the Z-direction. The height of the cones are about ½ of what they should be. Placing GCPs on the cones or anywhere on the rough surface is out of the question. I have tried adding GCPs to the sloped surface (the back part of the model) at varying heights, but this only makes the model reorient itself incorrectly with the X, Y, and Z axis.
 
My main question is: is there a way to get the model to correctly scale in all 3 directions using only scale bars placed on the back sloped surface?

My main thought would be figuring out the camera coordinate positions. Would this solve my issue, or would there be another way?

Thank you for your help!

Taber

[GPC]

This is not easy. You could "fake" GCP coordinates across the area using a simple tape measure.

For example, It you start at the tip of  one cone and call that point 1: located at 0,0,0 (X,Y,Z). Then measure a distance to other lower cones and estimating the up/down distance with a tape measure and level.

You would assume and X/Y orientation along the distances measured.

In other words:

point 1: located at 0,0,0 (X,Y,Z)
then measure a distance to another cone and assume that is in the Y direction, resulting in:
point 2: located at 0,delta Y, delta Z
point 3: located at delta X,0, delta Z

Not easy.. like I said. GCPs measured with a total station would be the best method.

[tabermiyauchi]

Thank you for your reply,

However, I'm afraid this will not work because the cones are only for validation purposes. The real surface that I'm modeling is much more stochastic in nature. Additionally, the object is too small to measure with a tape measure. From point 1 to point 2, is 1 centimeter.

I guess my main question is, if I knew the camera coordinate positions would this help resolve my issue with the z-scale scaling incorrectly?

[GPC]

If you knew the true values of the camera positions yes, it would improve the quality of your vertical results.. but at the scale you're working at, you would need to know the positions at the millimeter level. That doesn't say practical. I would recommend a LIDAR solution to modeling this small of feature.

[SAV]

Hi Taber,

If you are able to measure the back part of your model, then you place (tiny) GCPs there both at the bottom and the top of the slope and then use an arbitrary coordinate reference system. Don't forget to run the bundle adjustment (= OPTIMISE)  with this reference information enabled in order to improve camera self-calibration results.

Note: You could use a microscope stage micrometer to get accurate sub-mm length measurements.

All the best.

Regards,
SAV