add markers as Check points

[littlecloud47]

Hello,

I would like to use markers not to align the model, but only to use them to measure the accuracy from my model.
Is it possible simply to create markers, and insert their coordinates? So that I could measure the distance between (their new positions) them and the position they are on the images (the error).

[Alexey Pasumansky]

Hello littlecloud47,

The procedure is the same as for control points, but I suggest to add markers for check points after aligning and optimizing the cameras and, of course, such markers should be unchecked in the Reference pane after you load their coordinates.

[ThomasVD]

Hi Alexey,

So say you use coded targets (measured with for instance total station) to rectify the photogrammetry model + to control the accuracy. Then the procedure should be:

1. load images
2. detect coded targets
3. align images
4. load coded target coordinates (but uncheck perhaps 25% of the coded targets in Reference pane)
5. optimize sparse point cloud (using the 75% of coded targets still checked and which have coordinates)
6. control accuracy using the remaining 25% of coded targets not used for optimization

Is this correct?

[Alexey Pasumansky]

Hello ThomasVD,

Yes, that's correct. Also you can detect markers after camera alignment. In the latter case the projections of the markers wouldn't be used as valid matching points during the camera alignment.

[ThomasVD]

And detecting markers after alignment would improve the "correctness" of the accuracy control?

In the other case wouldn't detecting markers before alignment decrease alignment time and improve the probability of all images aligning?

[Alexey Pasumansky]

Hello ThomasVD,

That actually depends on the accuracy of the target detection. If the images are clear and targets are detected precisely, using detect markers option before alignment shouldn't affect the results. But if the images are blurred, noisy and the targets are not clearly visible, then the alignment results may be affected.
But since the optimization will still use markers, the difference will be minimal.

[ThomasVD]

Ok thanks for the clear explanation!

[ThomasVD]

Hi Alexey (and everyone else!)

Is there some advice you could give on which targets / how many percent of targets are best used to "optimize cameras" and how many are best used to check alignment accuracy? Is there some rule of thumb one could follow?

The recording scenario I have in mind is shown in the attachment: ca 50 targets have been spread more or less evenly across a table surface, and their locations have been measured using a Leica TCP 1205 Total Station. An object is placed on part of the table (so only targets in the direct vicinity will be detected in marker detection), the target locations of the detected targets (shown in red in attachment) are then loaded into the software.

=> which targets should be used for optimisation? 50%? 75%? Spread evenly across the table? Or simple the targets with the highest number of projections?
=> which targets should be used as accuracy control (and left unselected for optimisation)? How many percent? Should the ones with highest/lowest error / projections be preferred to get the most accurate representation of the maximum error? When reporting the total error of the model should only these control targets be selected, or should all detected targets be used?

Once again apologies for all the questions, but in short; the "PhotoScan workings" of accuracy control, error calculation, the meaning of different types of error (in PS Reports, the total error of GCPs is different than error of control points, which is again different than error of shown under "Survey Data") remain vague - if anyone knows of a good publication or forum discussion which clarifies these I'd be very happy to know!

Thanks,

Tom