Using fine registration (ICP) on riverbed data clouds

[fc64]

Hi,

I've been asked to register two data clouds : one photogrammetric point cloud from 2007 and one LiDAR point cloud from 2017. These clouds are already georeferenced and are centered around the riverbed. The precision of the two data sets is really different. The 2007 points spacing varies between 2 and 50 m (or more) whereas the planimetric precision of the 2017 cloud is 1m.

The area covered by the two entities is the alluvial fan of a river (see attachment file). This area changes a lot through floods and cyclones, and changed a lot between 2007 and 2017.

My goal is to evaluate the topographic evolution of the fan (sediment volume variation). According to some research articles a slight offset could impact a lot my volume calculations.


Here is my question : is it relevant to use ICP on these data sets even if the area changed a lot and the data sets have different precision ?

I tried to register the two cloud with a 10 % final overlap but the RMS is 1.73, it seems huge. I also tried 20% but the RMS is bigger. I'm wondering if using the ICP algorithm could create an artificial offset (which does not match the field reality).

Thank you in advance for helping me,

I hope my explanations are clear.

F.C

[daniel]

Hum, you could try, but you should use the newest (densest) cloud as reference in this case. And you don't need to reduce the overlap that much if the fan represent only 30 or 40% of the total area.

You could even clone the 2007 cloud and remove the fan with the scissors tool, then use this truncated cloud with ICP (90% of overlap?), and eventually copy the resulting transformation (in the Console) and apply it to the original cloud with 'Edit > Apply transfromation'.

[fc64]

Thank you very much for your answer Daniel.

I tried your different solutions but the RMS remains very high... I think the 2007 cloud is not dense enough to allow a good registration.

At least, I tried many solutions and I can justify why I prefer to use the initial cloud.