Modeling Thulamela

This page briefly highlights the modeling pipeline we used at Thulamela to build a 3D model. For a complete overview of Thulamela and an overview of its sociological context, please also visit this site

Under construction: New updates will come soon as the model is refined!!

Acknowledgements:This work was sponsored by NSF grant IIS-0121239


Click on image to view a short aerial clip of Thulamela

Thulamela is situated in the northern region of Kruger National Park, South Africa, and hosts the remains of a Great Zimbabwe civilization. Surrounded by ancient baobab trees, the stone walls of a major settlement have been recently rebuilt by the people of the local community following the ways their ancestors taught them. In late May 2004, our team travelled to South Africa to measure and build a 3D model of the site.

Team members:
      Prof. Peter Allen (1),
      Prof. Lynn Meskell (2),
      Hrvoje Benko (1),
      James Conlon(3),
      Edward Ishak (1),
      Benjamin Smith (1),
      Alejandro Troccoli (1),
      Lindsay Weiss (2)

(1) Department of Computer Science
(2) Department of Anthropology
(3) Department of Art History and Archaelogy

Working in KNP - May 29th - June 6th 2004

Equiped with a Leica HDS 3000 laser scanner, a Leica TPS 700 total station, and several still and video cameras, we acquired 3D point clouds, panoramic images, and video. Our aim: to combine all these elements and create 2D and 3D interfaces to navigate the site.

Building a 3D model

    These are the steps we took to build the model. We are still working on it, so check for future upgrades. Click on the images to view a larger version.
Point cloud data
The raw point cloud data consists of 28 range scans, and amounts to 40 million points. The HDS 3000 has a field of view of 360 degrees and a range of up to 200m, which allowed us to capture a course model in the course of 7 days. Scans can contain unwanted objects, mostly trees and bushes, that had to be removed manually.
Clean point cloud data
The unwanted objects have been removed. Trees and points outside the region of interest were eliminated.
Data partitioning
Due to its size, the data set was partitioned in smaller manageable boxes.
Surface integration
The data within each box was converted to a mesh using the VripPack (Volumetric Range Integration) software.

    View animation of the resulting model AVI (13 Mb)
Texture mapping
In addition to range data, the HDS 3000 can acquire 2D imagery. We used these images to create a textured version of the model.
Digital terrain model
Using the point clouds as starting point, we have created a digital terrain model of the site, to which we added texture from aerial photographs.

    View animation AVI (33 Mb)
Note If you have trouble playing the movie, you might need these codecs
Last updated: Apr 18 2005