 |
|
|
Wakulla 2 Post-Expedition Update
|
|
|
|
Wakulla 2 Post-Expedition Update
It's been four and one-half months since the last daily update and many people have asked what has been happening. After Mission Control was packed up and the scores of team members left Wakulla there remained the enormous job of processing the digital cartographic data collected inside the spring. One might expect this to be an automated process, but in fact it has proven very labor intensive. The reason relates to the availability of many independent data sets defining the internal 3D geometry of each tunnel (e.g. A-tunnel had ten separate data collection runs in which large sections of overlap occur). This can be viewed either as redundant data (to be thrown away or archived) or as a harvest of potentially ever-greater wall geometry definition in those zones where the overlaps occur.
Given the effort involved in running each mapping mission we have chosen the latter view in the post-processing of the data -- that is, to strive for the highest density point cloud definition of the cave wherever possible. This comes at the price of additional manual processing. The reasons are as follows. Prior to any mapping mission in a given tunnel radio location beacons were set and their positions on the surface were determined. These radio fixes formed a control system (known as the Wakulla Springs High Accuracy Reference Network or WS-HARN) for the subsequent 3D map. That is, when a diver operated the digital mapper they informed the onboard computer when they were over top one of the radio location points. In the final processing of the map, the coordinates for these radio location points become critical. They are the "hard points" through which the map must pass.
The Wakulla High Accuracy Reference Network
The first step in preparing an accurate map of Wakulla is thus the determination (on the surface) of the location of the waypoints. This process involved the use of four separate instruments (the custom-built magnetic axis detector; a Garmin differential GPS receiver; an infrared Leica total station; and a Trimble phase differential GPS site survey system). Each of these systems had their own coordinate systems and instrumental uncertainties. Throughout the spring and early summer we processed these overland survey data, developing both coordinate conversion routines, cross-checking position fixes of one technology against the other, and then computing an uncertainty statement for each point. It was essential that each radio location point have independent verification of the measured position and an uncertainty statement. These calculations took time. The result of this work is the WS-HARN Data Set and the WS-HARN Manual, both of which are now available on this website through the above hyperlinks. There are two specific things worth noting about the WS-HARN:
Vehicle Trajectories and 3D Wall Data
The remaining portion of this note addresses the 3D map data. With the control grid established the various mapping data sets must be fit through the control points. Although we know to a high degree of certainty where each HARN point is located we do not, in general, know exactly where the survey vehicle traversed between two consecutive points. This is the inertial measurement unit (IMU) drift problem discussed in detail in the HARN Manual and elsewhere on this website. In general we have found through calibrated tests that the free-floating data between control points can be brought back into alignment with the true trajectory through the use of a time-based least squares compensation algorithm provided the motion of the vehicle is relatively smooth and the turns are made relatively slowly. At Wakulla we used two further means to limit this drift since we knew it would not always be possible to achieve smooth, slow turns.. For example, we could space the HARN points closer together &emdash; as was done in B-Tunnel due to its innumerable twists and turns. We also were able to limit drift by zeroing out navigational errors at the start of each segment between HARN points (a process known as down mode alignment during which residual velocities and accelerations were reset to zero).
Even with all of these corrections if we were to plot all data collection runs going through two sequential HARN points we would see that the outlines of the points defining the walls did not match up precisely. This is due to the idiosyncrasy associated with each data run. In some early runs, for example, continuous passes were made over many HARN points without down mode alignment and thus error would continue to build as time accumulated. In many cases the outbound and inbound data sets differed, usually because of the water current slowing the into-the-cave run and speeding the exit run. In addition to the HARN points we had another very significant means of checking the alignment of the data: the sonar system, with its ability to resolve passage cross section geometry to +/- 1.5 centimeters, provided strikingly precise visual aids for registering otherwise free-floating data.
The procedure that has been presently adopted consists of first processing each mapping segment (of which there may be up to ten independent data sets from separate mapping missions) between two known radio location points. This brings all the data into relatively close proximity, but not exact alignment. Secondary processing then assigns a weighting function to each trajectory, based on both a visual assessment of the smoothness of each trajectory as well as discussions with the mapper operator regarding the specific conditions associated with each run. At this point auxiliary waypoints are taken into account. In general, for each radio location buoy located inside the cave the radio location crews also placed two additional, evenly spaced, intermediate buoys. By logging a flyover of these auxiliary points each of the independent vehicle trajectories was known to coincide at those locations, even though the exact geographical location of that point is unknown.
It then becomes a matter of using standard statistical processing procedures to adjust the map to the weighted mean position. This provides a best-estimate of the true location of the tunnel and wall points, and also provides an uncertainty statement for each point on the trajectory &emdash; something no dry cave survey has ever been able to achieve, let alone an underwater cave survey. Once this level of processing has been achieved there are still some discrepancies that are easily visible through the use of computer graphics &emdash; the human eye can, far faster than a computer, see that this bend in a tunnel, pocket in the wall etc is the same in all data sets and therefore must be registered. Because of the high precision of the sonar data it is possible to visually align these feature locations down to a few centimeters of error. We therefore have the basis for establishment of virtual waypoints which can be used in the same fashion as the physical auxiliary waypoint buoys inside the cave. It is this last operation that has proven to be a substantial time sink (just ask Barbara!) since it requires manual iteration to establish the best location for the virtual waypoints to bring each feature from each separate data run into exact alignment. The result, however, will be worth waiting for. Barbara has, through about 3 months of nearly full time work, completed processing of A, D, K, and O tunnels and is about 2/3 of the way through B-tunnel with C and F/L tunnels remaining. When the processing has been completed we will place an announcement on the web as to the availability of a CD-ROM containing the 3D wall point data as well as vehicle trajectories for each mapping segment. With some luck, this will be available by end of summer.
Bill Stone
July 22, 1999
Copyright ©1999, U.S. Deep Caving Team, Inc. All rights reserved. No portion of these pages may be used for any reason without prior written authorization.
