BY JACKSON COTHREN, ANGELIA PAYNE, ALEXEI VRANICH AND FRED LIMP

Laser surveying instruments (e.g., terrestrial light detection and ranging (LIDAR) and high-density survey (HDS) instruments) are increasingly becoming valued tools for geospatial professionals. They can quickly provide a dense set of 3-D data points that can be used to characterize buildings, engineering features and areas.

The spacing between data points is commonly less than 10 centimeters and can be as small as sub-centimeter (or, in some short-range cases, sub-millimeter). A particular challenge, however, is applying these methods over large distances that exceed 200 meters.

This feature discusses the use of an Optech ILRIS 3-D scanner in a challenging situation, a high-density survey of Machu Picchu, the famous Peruvian archaeological site. In summer 2005, with National Science Foundation support, researchers conducted an initial HDS at two World Heritage sites, Tiwanaku, Bolivia, and Machu Picchu, Peru, but this article discusses only the work at Machu Picchu.

An Unusual Site

Pachacuti, an Inca ruler, built the city of Machu Picchu in the late 1400s, high in the Andes Mountains outside the modern city of Cuzco, Peru. Machu Picchu was a place of worship as well as a royal retreat for Pachacuti and his guests.

Abandoned only decades after its construction, Machu Picchu was lost to the jungle, only to be rediscovered by Hiram Bingham in 1911. In the last few decades, it has become a major tourist destination.

When considering HDS sites as large as several hectares, such as Machu Picchu, laser-scanning systems that use time-of-flight (TOF) measurements are generally preferred. Wolfgang Boehler and Andreas Marbs review these as well as other scanning methods, including phase comparison and triangulation (see scanning.fh-mainz.de/scannertest/results300305.pdf).

The latter two techniques are limited in their range to typically less than 10 meters. And although the TOF systems provide greater range, they inherently yield less-accurate and sparser point clouds at longer distances than the shorter-range triangulation and phase-comparison systems.

Previously, TOF systems were frequently limited in their range to less than 100 meters. Now there are systems that can operate at more than 500 meters (and even 1,000 meters under optimum conditions). The Optech ILRIS 3-D unit currently supports one of the longest ranges of any terrestrial instrument.

Analyzing Laser Pulses

In a TOF system, a laser pulse is sent out, and a portion of the pulse is reflected from any surface encountered. The distance to the surface is calculated from the pulse's TOF. Precisely aligned mirror systems oscillate to control the horizontal and vertical angles of the beam, causing the pulse to move across the surface in a regular manner.

The mirror angles and distance to the target are used to calculate the X,Y,Z values for each pulse. Each system then produces a dataset that consists of multiple X,Y,Z,I values, where I represents the intensity of the reflected laser return. These points are sometimes referred to as a "point cloud."

Pulses often strike a surface at low, glancing (oblique) angles. This is particularly the case with large areas where the scanner can't be positioned significantly above the area and is aimed, more or less, perpendicularly at a surface. Furthermore, because the scanner can only measure surfaces illuminated by the laser, any surface that's obscured is in the "measurement shadow." Therefore, it's often necessary to perform multiple scans from different vantage points and integrate the results.

Such integration can be challenging. Scanning large areas creates large point clouds--numbers exceeding 200 million aren't uncommon. Commercial software systems, even those specifically designed for processing HDS data, are challenged to manage this many observations.

There's a large and growing body of literature about the effects of HDS scanner-pointing accuracy (including beam divergence), ranging accuracy and point spacing on the fidelity of the resulting point cloud. For the purposes of this discussion, all the effects can be aggregated, and the resulting "resolution" measures the level of detail that can be accurately extracted.

A facade with fine sculptural detail, for example, would likely require a much higher resolution than would a general architectural feature such as a stone wall, window sill or other feature with relatively simple geometry. Depending on the survey objectives, it may be necessary to use multiple scanners (e.g., TOF and triangulation) to accomplish all survey objectives.

Finding Facets

Although interactively viewing a point cloud can be effective, it's common to visualize the data by generating "facetized" (i.e., triangulated when the facets are triangles) mesh models, which are supported by many visualization and animation software packages. The resulting visual representation can be easier to interpret than a point cloud.

Facets can be produced from all the points in the cloud, but they're often generated based on a subset of the points, using sophisticated algorithms to intelligently "thin" the point cloud without sacrificing detail. In addition, many software packages provide tools that can be used to extract computer-aided design (CAD) elements from the point-cloud data and/or mesh.

Some software packages provide a variety of automated tools that extract edges and other features, but the process remains time consuming. If the scanning effort's final objective is to create CAD-level drawings, it will require substantial effort after the fieldwork. Research groups are working to integrate image-segmentation/machine-vision techniques into the CAD element-delineation process. If successful, these promise to dramatically reduce CAD feature-extraction times.

Scanning at Machu Picchu

Machu Picchu's layout is unique, and it presented several challenges. The site is located on the top of a steep mountain that has been carved to form a series of stone terraces and structures. The structures are tightly packed, and it's impossible to put a long-range scanner between them to obtain entire scans of each wall. A short-range unit could be used, but this would require a large number of scans and could have required months of collection and processing.

In the initial effort, most of the scans of were acquired from several distinct vantage points that provided panoramic views. As a result, the scans of Machu Picchu typically don't have the resolution necessary to resolve very fine structural details, but they show the general layout of the entire site at a respectable three-centimeter point spacing. The unique long-range capabilities of the Optech unit made this strategy viable.

The Optech laser scanner uses an eye-safe Class 1 infrared scanner, and it's possible to scan at any time of day or night. In the case of Machu Picchu, it was necessary to scan after the park closed, because there were too many tourists who would interfere with the scans during the day. Scanning typically occurred between 4:30 p.m. and 10 p.m. each evening, and each scan took about an hour.

Typically, three to four scans were acquired each evening, although this number increased as new methodologies for scanning at night were developed. The scanner was set up at six principal locations within Machu Picchu.

In addition to scanning Machu Picchu's ruins, scans also were acquired of the ruins on top of Huayna Picchu, a steep mountain to the north of Machu Picchu. Figure 3 depicts the scanning of Huayna Picchu, which was roughly 500 meters away from the scanner. The 3-D recording of this difficult-to-map structure was made possible given the scan unit's long acquisition range.

Scans of Huayna Picchu were acquired at three-centimeter resolution and took approximately two hours to obtain. A total of 25 scans were obtained across the entire site, resulting in a collection of more than 80 million data points.

Requirements and Implications

"Laser shadow" is a critical challenge for high-density surveys, and its effect should be clearly considered in field planning, particularly in large-area surveys. It's important to preview each scan to identify unexpected shadow areas in the field to ensure that all areas have been covered.

Where there's substantial relief or any structures present, it will almost always be necessary to have scans from opposite sides to ensure that areas aren't overlooked. If access to all sides of a location isn't feasible, it may be possible to set up the instrument in an oblique location and acquire the data. In the case of oblique scans, however, the geometry of the acquisition means that the resolution will vary more than it would with a perpendicular scan.

Scan Alignment

To provide a comprehensive dataset, it's necessary to merge the results (i.e., point clouds) of multiple scans acquired from different viewpoints. The survey of Machu Picchu, for example, required more than 20 scans from six locations. Merging implies aligning the scans by translating and rotating the point clouds until they lie in a common coordinate system. Only then can the scans be treated as a single dataset.

There are several techniques that can be used to align overlapping scans. These include identifying three or more well-defined targets that appear in both scans, using algorithms (such as Iterated Closest Point) that "find" similar data in multiple scans and "automatically" align the scan, and survey systems that precisely locate and orient the scanner itself in 3-D space.

Each of these techniques requires detailed planning to position the scanner so that all surfaces are scanned and enough overlap exists to align the individual scans. To help this task, most scanners integrate a digital camera that mimics the scanner field of view. Typically, 20 to 40 percent overlap is required to ensure proper alignment.

A further problem is error propagation through scan alignment. No alignment between two scans is perfect, and, as new scans are added to previous results, errors tend to accumulate. Therefore, it's common to "close" the alignment, much as a surveyor "closes" a traverse.

The first scan is aligned with the second, the second with the third, and so on. The last scan is made so that it overlaps with the next-to-last as well as the first scan, "closing the loop" on the dataset. Alignment software then can distribute alignment errors throughout the scans rather that have them accumulate in the last scans to be aligned. Although this type of closure isn't always possible, it should be considered. The work at Machu Picchu, for example, used PolyWorks 9 from Innovemetric (www.innovemetric.com) to process, align and analyze the point clouds.

Resolution and Range

Although each measurement by the ranging laser in a TOF system represents an infinitely small point, the beam itself has a finite diameter. The measured "point" is a function of the integrated return pulse over the area that the laser illuminates. As a result, resolution is, in part, a function of the beam's diameter.

As the beam travels, it diverges, so the potential resolution decreases as range increases (this is only one reason for reduced resolution at long ranges). The formula provided by Optech for the ILRIS-3D beam width as a function of range is the following: diameter in millimeters = 0.17 * (range to target in meters) + 12 millimeters.

Thus, beam diameter at 500 meters is 97 millimeters. The spot-size effect is shown graphically in Figure 4, where spots of various sizes are superimposed on a surface, with reference points spread across the surface at a nominal 20 millimeters.

The impact on beam size and resolution is important but complex. For a distance measurement to be made, only a portion of the pulse needs to be reflected from the surface to the instrument. As the distance becomes greater, the strength of the pulse is dispersed, and the returning energy level is similarly reduced. Highly reflective surfaces return more energy, so it's possible to measure them at greater distances than less reflective ones.

For Optech systems, the instrument can operate at a range of 350 meters from a surface that's only 4 percent reflective. At 800 meters, the surface needs to be approximately 20 percent reflective or better. Each instrument will have its own properties, but this general pattern will always apply.

Data Distribution

A common limitation of HDS is the difficulty of data distribution. For example, there's currently no standard data format, as there is for airborne LIDAR. An important step in increasing the distribution of HDS data, however, has been the release of a free large-point-cloud data viewer, IMView, by Innovmetric (see www.innovmetric.com/Manufacturing/viewer.aspx).

With IMView, it's possible to distribute point-cloud data that recipients can analyze, determine a variety of measurements and investigate the data. It's possible to measure distances and angles as well as define a range of geometric primitives, such as cylinder radii. It's also possible to define cross sections, develop annotations, output reports and perform several other actions.

Using these tools, HDS data from Machu Picchu and Tiwanaku were placed on the Web at www.cast.uark .edu/invirmet. The site also links to the IMView software along with tutorials on using IMView with the Machu Picchu data.

--------------------------------------------------------------------------------

Authors' Note: Material in this article was revised and abridged from material in a forthcoming International Council on Monuments and Sites publication. The Machu Picchu survey was supported, in part, by NSF IIS 0431070 and NSF BSC 0321286 research grants. Work at the site was made possible through the assistance of the Kuraka Association and the Instituto Nacional de Cultura Cusco. Innovmetric provided an additional license of its Polyworks software for field use in South America. Dean Collis Geren and Jack Ezell assisted with international equipment transfer. We'd also like to thank JR Payne and Snow Winters for essential in-country assistance.

--------------------------------------------------------------------------------

Jackson Cothren is an assistant professor and researcher with the Center for Advanced Spatial Technologies (CAST) at the University of Arkansas; e-mail: . Angelia Payne is a CAST researcher; e-mail: . Alexei Vranich is a University of Pennsylvania museum research associate; e-mail: . Fred Limp is a professor of Geosciences and Anthropology, and director of CAST; e-mail: .