A bit of warning: this is a rather technical description which serves as experiment documentation and is not intended for general public. It is still incomplete and will be updated. Any comments are welcome.
The experiment has shown that as many as 58 partial views can successfully be registered with high precision in a common coordinate system without any prior knowledge of the camera motion parameters if suitable landmarks and simple registration aids are used.
The goal of this experiment was not to obtain any sort of surface (manifold) representations. That is the topic for our ongoing research.
There is no limitation on the number of partial views that can be processed (except for computer disk space and processing time, which both scale linearly with increasing number of views).
The wider field of view of the b/w cameras was necessary to provide some view overlap necessary for successful view registration. Only the b/w cameras were used for reconstruction, the color camera was only used for "coloring" the reconstructed points. In the color camera AGC was switched off to get consistent colors over all partial scans.
The dataset (the total of 904 b/w and 60 color images) was processed on a SUN Sparc station with 250 MB of main memory.