Traversing Complex Environments Using Time-Indexed High Dynamic Range Panoramas
Static panoramic photography has been shown to contribute to
context-rich descriptions of regions under archaeological study
[Allen et al. 2004]. We show that fast traversal through a matrix
of dynamic panoramas can allow users to quickly locate specific
target features within a complex scene. Results are presented
using two large archaeological monuments as a test subjects.
2 Capture and Processing
We collect source images using a camera with a circular fisheye
lens, articulated by a custom rotation controller. Capture is
automated. The camera is positioned by unidirectional,
monotonic rotation around the lens nodal point. Exposures are
taken at ten discrete rotation intervals. In order to enable high
dynamic range output, three bracketed exposures are acquired for
each camera position.
Figure 1. Initial image assembly from fisheye images.
As the capture process iterates, a time-lapse sequence is built
frame by frame with each successive revolution of the camera.
The rotation period is approximately 1.8 minutes. In the
processing steps, source images are geometrically calibrated to
account for lens distortion and an HDR representation is
computed. A rendered view for one frame of a panoramic
sequence is shown in Figure 1.
In our viewer application, we first index the recorded panoramas
in a spatial network, using global coordinates recorded during
image capture. After initialization, users interactively navigate
through the image data using a novel multi-node viewer in which
each node represents a different location on the time axis. The
user’s current position is drawn in the viewer’s main window.
The 3D location for each image capture location in the dataset is
projected in this view as navigation links (Figure 2). By hovering
over a given link, the user obtains an interactive preview of the
linked panorama. In a separate window, a ground plan of the site
enables spatial domain traversal of the scene. Indexed metadata is
presented in a third window. View information between windows
is updated in real time via an XML data stream (Figure 3).
Figure 2. User interface for the panoramic viewer.
2 Results and Validation
To test our system, we acquired 30 time-lapse image nodes in situ
at Chichén-Itzá, Mexico. Separately, we acquired 60 image nodes
at the Temple of Ramses II, Egypt. Archaeological researchers
were asked to use our viewer system to identify 20 specific, non-
repeating visual features present in each actual scene. They were
then asked to identify the same features by viewing the set of
rectified input photos gathered during our capture step. We found
the seek time for a given target feature when using our system was
faster by a factor of 3.1 for the smaller data set, and a factor of 7.4
for a larger data set. Our approach was shown to be an efficient,
low cost technique for interrogating complex real-world scenes.
We also found that the time advantage scales with database size.
Figure 3. Components of the interactive viewer
ALLEN, P., FEINER, S., TROCCOLI, A., BENKO, H., ISHAK, E. AND
SMITH, B. 2004. Seeing into the Past: Creating a 3D Modeling
Pipeline for Archaeological Visualization. In Proceedings of
the 3D Data Processing, Visualization, and Transmission, 2nd
international Symposium on (3dpvt'04), IEEE Computer