Aqua Phoenix
     >>  Research >>  Beauvais  


2. Data

2.1 Cloud Point Scans

A 3D cloud point scan is a collection of points in 3 dimensional space that, in the case of this project, represent samples of the surface of as building structure. The collection of cloud points in this case has been retrieved with a 3D range scanner. Because a range scanner is a stationary device with a limited range in the horizontal and vertical directions as well as in the distance of the projected beam, a large structure requires a number of individual scans that need to be registered (stiched together) in a later process.

Cloud point scans for this project tend to contain on the order of 100,000 to 1,000,000 points. Altogether, there are 121 scans from the inside of the cathedral, all of which combined contain more than 50 million points. In a standard uncompressed XYZ format, this takes about 2.5Gb of storage; Compressed, this number reduces to about 900Mb.

As there are no definitive and accurate methods of registering cloud point scans algorithmically, the first stage of registration, as performed in this project, involves registering scans manually, so as to get a first estimate of the structure. In this estimate, the point cloud scans are roughly in the correct position relative to all the other scans. From there on, the scans can be registered correctly using algorithms.

Manual registration for the 121 cloud point scans has been accomplished by using the software package EvalViewer by Alias|Wavefront. While this software package has been used throughout the entire registration process, it is not ideally suited for as large a data set as was necessary for this project. It turned out that on a Sun Blade 1000, 2x750MHz, 1Gb RAM the maximum number of points that could be loaded at any given time was around the order of 10 million. For the purpose of registering all of the 121 scans, scans were first registered in smaller groups, up to 5 million points). Using a skeleton of scans that roughly covered the entire cathedral, the groups were then registered to this skeleton one at a time. Needless to say, any small errors within the groups then multiplied to generate larger errors in the final global registration.