Tuesday, September 20, 2011

Virtual Polynomial Texture Mapping, Structure from Motion, and Pole Aerial Photography at the Guadalupe Village Site (LA 143472)

3D rendering of Feature 59.
Here are initial results of a mapping exercise using a combination of inexpensive and innovative technologies.  As part of a larger project, the mapping techniques were tested on a prehistoric burned rock midden (Feature 59). The feature, sometimes referred to as a ring midden by local archaeologists, measures about 18 meters in diameter and is more than one meter tall.  It is just one out of more than a hundred burned rock features found at the Guadalupe Village Site in southern New Mexico.
Feature 59 in August 2010 after a very wet spring.
Feature 59 in July 2011 after wildfires hit the area.

We first aerially mapped the site in the summer of 2010 but had mixed results due to usually dense vegetation that had grown over it, following an uncommonly wet spring season. Wildfires burned much of that brush off in 2011 at which time my colleagues and I were invited back to the site for another go at it.

The focus of the main project was to document the site with Kite Aerial Photography (KAP) and Blimp Aerial Photography (BAP), though the data highlighted here were collected from a handheld pole.  A makeshift Pole Aerial Photography (PAP) rig was cobbled together from a painter's telescoping pole, a modified paint roller, a "Tupperware" container, few zip-ties, and electrical tape.  The rig allowed for a Canon A540 digital camera to be pointed straight down while suspended more than five meters above the ground.  The camera was programmed to automatically take photographs every several seconds using the Canon Hackers Development Kit (CHDK) while running an intervalometer script.
Conducting PAP at Feature 59.
After clearing some of the burnt plants away from the feature, the PAP rig was slowly walked across the it in a series of transects.  The objective was to take a number of overlapping photographs across the entire surface of the feature.  184 photos were collected in this fashion and, after setup, took less than an hour to complete.

Example of overlapping images collected during PAP.
Once back from the field, the number of photographs was culled by removing blurry and off-subject images.  As a result, 158 photographs were processed using Structure from Motion (SfM) techniques and a textured high resolution 3D model was created.

Recently, I tested Polynomial Texture Mapping (PTM), a type of Reflection Transformation Imaging (RTI), to enhance petroglyphs in West Texas.  I wondered if the same process could be applied, in a virtual sense, to the 3D model of the feature.  To over-simplify it a bit, the PTM process involves taking a series of photographs (normally under dark conditions), while an off-camera flash is moved around the subject matter from set distances and angles.  The images are then imported into a piece of software that allows the object to be examined as a polynomial representation of all the photographs combined.  PTM often reveals details that are not visible to the naked eye and can even brings out structural elements not visible with laser scanning.  Using this technique we are able to look at the form of the Feature 59 in a new way.

Creating the VPTM Model
Virtual lighting and camera setup in 3dsMax.
Virtual PTM renderings.
PTM of 3D model in PTMviewer software.
To create the PTM model, a set of virtual lights were positioned around the 3D model in 3D Studio Max.  A total of 144 lights were stationed in dome-like fashion around the feature.  A virtual glossy sphere was also created and placed next to feature.  The sphere is used by the PTM building software as a reference for the location of each light source.  Using this virtual photography studio, 144 images were rendered, one for each of the lighting locations.  Finally, those images were fed into the PTMbuilder application through a java interface called LPTracker and a PTM file was created.
PTM rendering with possible sub-features highlighted.
The Virtual PTM (VPTM) reveals many aspects of the feature that are not visible in the textured model. It suggests that sub-features may have been excavated into the main feature or, at least that burned rock was removed from it at several places. Some of the possible sub-features are obvious when standing in front of the actual feature but, not all.
Generating a PTM is this fashion is particularly innovative because it would be almost impossible to do with the real life feature.  Two cranes, one for the camera and another to move around a giant light, would have been needed to create a PTM of Feature 59 using the traditional approach.  The use of VPTM on objects that do not allow for the systematic photography that a normal PTM requires, may reveal aspects that are impossible to see otherwise.  Using VTPM with aerial photography is just one possibility.  For example, the same process could be used on hard to reach rock art sites or within submerged caves.

Creating a VPTM from a 3D model has some pitfalls in that the VPTM can only be as good as the 3D model it is generated from.  Any distortions in the 3D model that are an artifact of the SfM process could appear as prehistoric anomalies in the rendering.  Researchers should consider this when making and evaluating similar models.

The feature documented here, used equipment that cost well under $250 (most of that for the camera).  The PTM generating software is free. While the other software used is more expensive, free open source solutions capable of the same results are available.  Considering the minor amount of time needed to collect the data, the low cost of the process, and the high quality results, other archaeologists should consider applying these techniques to sites anywhere they work.
Digital Elevation Model (DEM) of feature with 5 cm contours. Note that west is up.
The initial work at the Guadalupe Village Site (LA 143472) was funded by a small Permian Basin MOA Grant through the Bureau of Land Management office in Carlsbad, New Mexico.  The second visit to the site (which yielded this data) was done entirely pro bono. Archaeologists Juan Arias, Bruce Boeke, Tim Graves, Jeremy Iliff, and Myles Miller III made the project possible.