What is the registration error box under the M3C2 parameters actually doing? And what happens to the output if you change the numbers? I have been changing it from .01 to .05 depending on my error from another program. Should I being leaving it set at 1.0?
Thank you,
Carly
Registration error in M3C2 tab
Re: Registration error in M3C2 tab
If you know the registration error (when you registered the two clouds you compare), then M3C2 can take it into account during the process and notably when computing the distances confidence.
Daniel, CloudCompare admin
-
ilovecoffee1125
- Posts: 36
- Joined: Fri Oct 14, 2016 9:00 pm
Re: Registration error in M3C2 tab
Daniel,
What would you recommend this number to be? And what does registration error mean in cc?
Thank you.
What would you recommend this number to be? And what does registration error mean in cc?
Thank you.
Re: Registration error in M3C2 tab
Hi, it depends on your both point cloud quality. Reg = sqrt(sigmapcloud1^2+sigmapcloud2^2). It is important for significant or not significant movement decision after comparison. You can read paper about M3c2.
Cheers
Cheers
Re: Registration error in M3C2 tab
Hi,
the registration error is independent of the point cloud roughness (well, except if you base your registration on ICP for instance...but that's a different problem...).
To quantify the registration error, there are various ways:
. either you have parts of your two clouds that you know have not changed at all, and you evaluate what is the distance between these parts. If you find a large distance, you should try to improve your registration so as to minimize the average distance between your fixed elements (e.g., playing with ICP or your registration software).
. or if you register your point clouds with independently known target, the registration can be derived from the rms of the registration (it's not directly the rms, it's a bit more complex than that as it depends on the registration quality of each cloud to the target).
If you don't have any idea, you can leave it to 0, but you'll have to remember that the confidence interval predicted by M3C2 and the statistical change will not account for registration error. The confidence interval would be a minimum estimate. Depending on the type of data you're using and registration method, and type of surface (smooth or rough), the registration error can completely dominate the error budget of the confidence interval, or be negligible... In the end it also depends on the amplitude of the change you want to detect. If it's not really different from the registration error, you'll have difficulties to assess that your change is statistically significant. If the change is much larger that the expected registration error (e.g., 50 cm change, when you expect 2-3 cm registration error (as for e.g., good airborne lidar data)), you don't really need to bother with the registration error.
Cheers
Dimitri Lague
the registration error is independent of the point cloud roughness (well, except if you base your registration on ICP for instance...but that's a different problem...).
To quantify the registration error, there are various ways:
. either you have parts of your two clouds that you know have not changed at all, and you evaluate what is the distance between these parts. If you find a large distance, you should try to improve your registration so as to minimize the average distance between your fixed elements (e.g., playing with ICP or your registration software).
. or if you register your point clouds with independently known target, the registration can be derived from the rms of the registration (it's not directly the rms, it's a bit more complex than that as it depends on the registration quality of each cloud to the target).
If you don't have any idea, you can leave it to 0, but you'll have to remember that the confidence interval predicted by M3C2 and the statistical change will not account for registration error. The confidence interval would be a minimum estimate. Depending on the type of data you're using and registration method, and type of surface (smooth or rough), the registration error can completely dominate the error budget of the confidence interval, or be negligible... In the end it also depends on the amplitude of the change you want to detect. If it's not really different from the registration error, you'll have difficulties to assess that your change is statistically significant. If the change is much larger that the expected registration error (e.g., 50 cm change, when you expect 2-3 cm registration error (as for e.g., good airborne lidar data)), you don't really need to bother with the registration error.
Cheers
Dimitri Lague