The pipeline of co-registering laser scans and images

Hey mates

Currently I am working on my MSc thesis which is aimed to study the potentiality of fusing laser scans and digital images for cultural heritage documentation.
I am using STONEX X300 laser scanner and Nikon D5200 camera (18mm). JRC 3D Reconstructor (Provided by STONEX) was used to register scans to be exported then as E57 format scans (which is supported by RC).
Here, I am listing my initial trials on RC, I need to know what is going on behind the scenes, scientific explanations, the meaning of options available on RC environment. Any participation will be appreciated!

Object: The Assyrian Winged Bull
No. Of Laser Scans: 7 Scans (Imported as 11 laser images on RC)
No. of Images: 69 Images

1. I have imported the scans (E57) with the following settings: Registration= Exact, Georeferenced = True & Prior pose (Absolute pose) = Locked, but they were misaligned. I tried changing the absolute pose to other options listed there but it didn’t work too !

2. I imported the scans (E57) with the following settings: Registration = Draft, Georeferenced = True & the Absolute Pose = Unknown, only 5 scans (out of 7) have been aligned and the Median and Mean errors are 1.99 and 1.05 respectively! ( see the image)
Changing the Absolute Pose = Locked, just turned the model into a mess! And other Pose options are similar to the Unknown where errors are over the limits.

 

3. I returned to the JRC (STONEX) and exported the scans as ( UCS-E57 ) and imported them to RC with the following settings: Registration = Exact, Georeferenced = True & Prior pose (Absolute pose) = Locked, and they aligned but the Median and Mean errors are ZEROS! ( is it alright to be zeros ?) (Points count = 232391 )

 

My trials to integrate lasers and Images: 

Firstly I imported the 69 Images and they were aligned perfectly with settings recommended in the forum by Wishgranter (thanks to him) and the Median and Mean errors = 0.5 and 0.6 respectively. (Points count 417572 ).


Then I Imported the 7 scans (UCS-E57) and placed some control points ( tie points ) on Images and laser images aiming to enhance the quality of co-registration between the two entities but when I aligned them all points were indicted to have a projection error over the limits (the re-projection error was set on 2) so they canceled out and the alignment proceeded to produce a combined point cloud with 351845 points count which is less than the one generated by images only! The Median and Mean errors = 0.5 and 0.6 respectively.


Finally , I deactivated the CPs and re-aligned the components to have 582035 points count and the same errors (0.5 &0.6 )

So any ideas or comments which in turn may clarify the issues I faced and how to assess the accuracy of combining laser scans and images ? 

Thanks in advance 

  • Abbas Sh.