points, lines, surfaces, features models from point...

I

HIGH-DEFINIT ION SUR VEYING: 3D LASER SCANNING

Geoff Jacobs

Points, Lines, Surfaces, Features Models from Point Clouds

In an earlier, introductory arti-cle on high-definition survey-ing (3D laser scanning), it waspointed out that its single most

important, distinguishing feature is itshigh point density (May 2004). Otherkey features—high-speed data capture,reflectorless measurement, 3D visuali-zation, and colorful imagery—can beconsidered bonus by-products of thetechnology’s focus on economically cap-turing a site or structure with high pointdensity. The potential advantages ofdense point data, such as creating morecomplete and accurate as-builts, havealso been covered in earlier articles(May, July, and August 2004). High pointdensity, however, raises some interest-ing new questions—and opportuni-ties—regarding the task of extractingtraditional deliverables.

What Do I Do with All Those Points?

Whether traditional surveying is donewith a tape or chain, a total station, orGPS, the focus is on the collection of indi-vidual, significant points. In a day’s work,hundreds or even thousands of points ofinterest are collected intelligently. A pro-fessional places a pole/prism on, or aimsa reflectorless instrument at, selectedspots to be measured by the instrument.Surveyors intelligently collect points (andsometimes add codes to identify associ-ated features) at markers and monu-ments, corners, intersections, edges, se-lected features, etc., knowing that theseselected points/features will form a basisfor creating final deliverables.

In contrast, for the same site or struc-ture, a high-definition or laser scan sur-vey can literally blanket it with millionsof closely spaced points. This article willdescribe various ways of extracting,from high-definition survey data, the de-sired points, lines, surfaces, symbols,models, and other elements needed forfinal deliverables. This article will alsodescribe an increasingly popular newdeliverable option—point clouds them-selves—and how progressive clients to-day are effectively using this deliverableto better perform their engineering,planning, QA, analysis, archiving, andother tasks.

Loading Point Clouds for

Processing

Data captured by laser scanners isstored on electronic media. Data is typi-cally stored directly on a portable com-puter, along with software that operatesthe scanner and may also be used toview and process the data. Dependingon the scanner, data may alternately bestored on a removable media/card thatresides either on the scanner or in ahandheld collector, but without 3D view-ing and processing capabilities.

Some vendors provide both scannerhardware and point cloud processingsoftware; others supply one or the oth-er. Point cloud software is either file-based or utilizes a database structure,which can be beneficial for managinglarge data sets and for sharing scan da-ta among multiple users. Most pointcloud processing software can import

standard file types, such as simple ASCIIformats, but each software has an inter-nal, native representation on which itoperates most efficiently. Vendors thatprovide integrated scanning hardwareand point cloud processing softwaretypically make it such that no extra dataconversion is needed to process the out-put from the scanner. That is, Vendor X’sscanner outputs data in a format that isoptimal (native) for Vendor X’s software.When mixing and matching is done be-tween point cloud software and scan-ning hardware from different vendors,an extra step of data import and dataformat conversion must be done. Theconversion time depends on thesize/format of the files and the horse-power of the processing PC.

Creating Registered,

Geo-referenced Point Cloud Sets

The first step for creating deliverablesfrom any high-definition survey is to regis-ter together and/or geo-reference pointclouds captured from different scanner

Multiple scans registered and geo-refer-enced. Image courtesy BSC/Cullinan, Inc.

&ExtractingExtracting &

DISPLAYED WITH PERMISSION • PROFESSIONAL SURVEYOR MAGAZINE • September 2004 • WWW.PROFSURV.COM • ALL RIGHTS RESERVED

setups. Even for “single-scanner-position”projects, it may be important to tie scandata to site coordinates. Today, variousways are available to register and geo-ref-erence point clouds. Some scanners canbe directly geo-referenced over points,similar to the way a total station is used,or can utilize GPS. For high-accuracy worktoday, the common registration/geo-ref-erencing methodology for scanning isanalogous to that for photogrammetry:known points or targets within the scansare used for registration/geo-referencing.In addition, overlapping scans may alsobe aligned together. In many cases, bothknown points and overlapping scans areused to create a final “registered, geo-ref-erenced” point cloud data set.

Data Cleanup

A second step sometimes used onhigh-definition surveys is cleaning upthe data, i.e., removing unwanted dataprior to creating final deliverables.

In high-definition surveys, unwanteddata can come from various sources. Forexample, laser scanners can literallycapture everything within range of thescanner, including people or vehiclespassing through the scanner’s line ofsight. Likewise, not all static featurescollected (e.g., vegetation, constructionjunk, tools, etc.) may be pertinent to thedeliverable. Additionally, laser scannerscan collect spurious data reflected offmirrored surfaces (including puddles ofwater). Yet another source is “edge ef-fects,” where a laser beam that partiallyreflects off the edge of an object may

give a “false point” that appears by it-self out in space. The edge effect phe-nomenon worsens with increasing spotsize and is often more pronounced withphase-based scanners, which rely on in-tegrating consecutive point samplings.Phase-based laser scanners are alsovery photon-sensitive and can recordambient light as random point noisewithin a scan. They are also subject tophase ambiguity resolution effects,which may require additional data clean-up (an explanation of this is consideredbeyond the scope of this article).

In practice, scan data from phase-based scanners is filtered in the office toremove most noise. Users report good re-sults for this process, even though 25%

or more of the data collected may be fil-tered out. For time-of-flight scanners, da-ta may optionally be cleaned up before itis used, depending on required deliver-ables. I’ve spoken with experienced time-of-flight scanner users who used to cleanup their data sets before processingthem, but now don’t bother with that stepif the extraneous data does not negative-ly impact results or office efficiency.

Extracting Single Points

Each point in a point cloud hasunique X, Y, Z coordinates, so a userneeds only to click on a point to observeits coordinates. This can be done in spe-cialty point cloud software or in popularCAD applications such as AutoCAD orMicroStation (or in applications built ontop of these platforms). For point extrac-tion within CAD, there are two possiblepaths. One path is to import the pointcloud or part of the point cloud into CAD.This path is efficient only with relativelysmall scan data sets (in the low tens ofthousands of points). The other path isto use special point cloud “CAD-integra-tion” software, much like a CAD plug-in,that allows large point clouds to be ac-cessed from within CAD. Such softwareis commercially available for both Auto-CAD and MicroStation and is being de-veloped for other CAD applications.

As easy as it is to extract singlepoints from point clouds, you can’t besure that a given scan point is exactly ona corner or an edge, or represents the

Virtual surveying: scan points and codes can be exported to mapping software.

Users can click on any point within a point cloud for its X, Y, Z coordinates.




lowest point of a curb, etc. Use of a sin-gle scan point in this way will depend onthe accuracy required. If a user onlyneeds point accuracy on the order of thescan spacing, say 2cm, then it may beacceptable to use single points from thescan data, depending on the actual sin-gle-point accuracy of the scanner. Ifhigher accuracy were needed, then analternative (again depending on the ac-tual single point accuracy of the scan-ner) would be to execute a high-densityscan of the area of interest (e.g., with1.5mm point-to-point spacing) and clickon the point of specific interest.

If even higher point accuracy is need-ed, then it may be necessary to create amodel of the area of interest and derivethe point of interest from the model. Thisis exactly the process that is used to lo-cate the center point of scan targets usedfor geo-referencing. A planar or spherical

target is tightly scanned (e.g., 1.5mmpoint-to-point spacing); fitting algorithmsare then applied to fit the target, repre-senting the center as a vertex. Target cen-ters acquired this way can be derived to1.5mm accuracy, depending on the typeof scanner used. Most software allowstarget centers to be extracted automati-cally. A similar approach can be appliedto other objects/features. This techniqueof doing fine scans to extract highly accu-rate, single point features is commonpractice today, provided that the scannerhas this type of capability.

Extracting Point IDs, Features

and Symbols

When professionals first see pointcloud data, they may quickly note that apoint cloud has unmistakably captured atraffic sign or light pole or other objectsoften mapped as symbols. The pointcloud provides the highest level of confi-dence of what the object is and where itis; the question is, “How do I map a fea-ture that I see in the point cloud as a sym-bol in my deliverable?” Today, software al-lows a user to select a point from the pointcloud and attach a point ID with standardor custom feature codes to that point.This is essentially “virtual surveying,”whereby point cloud software emulatesthe functionality of standard handhelddata collectors, except the work can hap-

pen in the office rather than in the field.When the selected point and featurescodes are exported via ASCII format andExcel tables to standard mapping soft-ware, the symbols are likewise recog-nized and placed on the map deliverable.Some software also provides automaticsequential numbering capabilities.

Taking Distance Measurements

All commercially available pointcloud software has the ability to extractslope distance measurements from anytwo selected points. This capability isvery useful for quick, approximate dis-tance measurements. Horizontal andvertical distances can also be readily ex-tracted, provided the software has thiscapability. Again, the accuracy and utili-ty of the distance measurements fromthe point cloud are constrained by thecloud’s point-to-point spacing and the ac-curacy of each selected point. For moreaccurate distance measurements, pointcloud modeling is needed. As notedabove, such distance measurements canbe made within specialty point cloud soft-ware or directly within CAD.

Extracting Lines

Creating line work is a mainstay ofsurveyors and other measurement pro-fessionals, and it is here that many won-der how you get from dense point cloudsto traditional line work, especially 2Dplans and elevations. In the earliestdays of laser scanning, the only way toget to 2D plans and section line workwas to first create a 3D model and thenextract 2D plans and sections from the3D model. Today, however, there aremany ways to get to 2D (or 3D) line workwithout having to create any 3D models.

Distance measurements to points and surfaces

Section created by tracing over scan datausing point cloud CAD-integration soft-ware and MicroStation. Image courtesyWashington Group International.

Roadway cross-sections at regular Intervals


Choosing the best method for line gen-eration is dependent on how accurate theline work needs to be and how efficiently itneeds to be generated. One of the fastestways to generate line work from pointclouds is to simply trace over points withinthe cloud. “Digital tracing” is simply the actof moving the cursor along the point cloudand clicking on or near individual points ina sequence. Line work is then the trail ofthe digital tracing. If the operator wants toensure that only certain points are select-ed, the operator can zoom in on the pointcloud area of interest. Software tools arealso available to isolate portions of pointclouds for doing such work.

Using visualization aids and/or slicingtools, it is fairly easy to get a view of thepoint cloud, either in 3D or 2D, whichclearly shows where to trace over thepoints. Some software tools automatical-ly adjust the manual tracing to best fit astraight line, a common task for creatingline work of buildings. Some software willalso automatically create line work frompoint clouds without intervention fromthe operator, but this approach appearsto have a long way to go before it is ro-bust enough for creating final deliver-ables. The problematic issues here arethat fully automatic line generation re-quires high point density and regular ob-ject geometry. In the real world of laserscanning, adequate point spacing andregular object geometry are not alwayssimultaneously present. Many architec-tural facades, for example, have highly ir-regular geometry that make automaticline fitting quite difficult.

Another usefulmethod for creatinglines is “semi-auto-matic” line generation.In this case, the opera-tor focuses on a specif-ic region of the scandata with a prioriknowledge about thetypes of geometry thatare applicable. Goodexamples of this arethe creation of linework for curbs andgutters and the cre-ation of “line work”

representing road cross-sections at regu-lar intervals. [Although road cross-sectionsare not commonly referred to as “linework,” the sections are lines—as opposedto points or surfaces]. For curbs and gut-ters, a “cross-section template” can becreated that best fits one section of thecurb/gutter. Once this template is created,the operator can use this template to au-tomatically follow curb and gutter points

3D line work from point clouds. Image courtesy Bohannan Huston, Inc.

Surface created from high-definitionsurvey. Image courtesy BSC/Cullinan, Inc.

3D model of historicbridge, created fromhigh-definition survey.Image courtesy Bohannan Huston, Inc.

that are continuous within the scan datathus creating line work. In the case of roadcross sections at regular intervals, the usercan select large portions of the point clouddata set and then specify within the soft-ware the intervals at which sections are tobe generated. The resulting sections aregenerated within seconds ready for exportto CADD software.

Contour lines can also be generatedautomatically from surface meshes creat-ed from point cloud data. Data cleanup isimportant when generating the originalsurface meshes; otherwise contour linesmay not follow the surface of interest.

Extracting Surfaces

Many types of point cloud softwarehave the ability to generate surfacemodels. These can be created frommeshes or from CAD primitives, whereappropriate. For civil applications suchas road surfaces, initial work is donewithin the point cloud software tocleanup the data and to create TINmeshes. Meshes can be decimated intel-ligently to reduce file size while retain-ing critical changes in geometry. Afterthis step, the points themselves are of-ten exported to civil design software forcreating final DTM deliverables.

Some software provides the capabilityto create NURBS surfaces, which arecomplex mathematical models of irregu-lar surfaces. This type of software is com-monly used in mechanical engineering(e.g., automotive parts and reverse engineering) and in “organic modeling”domains for the entertainment industry




click on a point that obviously lies on thesurface of a pipe (for example) and in-struct the software to fit a cylinder usingall other appropriate points near the se-lected point. Similarly, the user may selecta point on a wall or floor and instruct thesoftware to create a plane, and so forth.Tools allow the management of CAD-stylelayers so that all piping can be shown inone layer, all equipment in another layer,etc. Three-dimensional models can be ex-ported to intelligent 3D design softwarethat enables the creation of “intelligent3D models” that represent additional at-tributes of the object, such as the thick-ness and pressure rating of piping.

Three-dimensional models of irregularobjects, such as statues, cars, boulders,etc. can also be created, but this involvesmore intensive office processing and theuse of specialty software. Likewise, tex-ture mapping can be applied to the pointclouds or geometric surfaces for photo-realistic models, but this is also generallydone outside of point cloud software us-ing other specialty applications such asAutoDesk’s 3D Studio, Bentley’s TriFor-ma, etc. As with complex surface model-ing, creating photo-realistic 3D modelsmay involve competencies not commonto civil/survey organizations.

Using Point Clouds Directly – the

New Survey Deliverable

As noted above, there are many waysto create a variety of deliverables frompoint cloud data sets. Perhaps the mostinteresting development, however, is theemerging use of point cloud data sets as adeliverable, without extracting anythingfrom the data! Here’s the scenario: A pip-ing designer needs to check if the pro-posed routing of a new pipe through anexisting plant will clash with any existingpiping, structures, equipment, stairs, etc.Rather than develop an as-built of the ex-isting plant and then import the as-builtinto the piping design model, the design-er can simply integrate the proposed de-sign with the point cloud. Software toolscan automatically detect any interferencebetween the proposed design and thepoint cloud. Measurement tools can beused to determine the geometric extent ofany clash and the proposed design can beadjusted and re-checked.

The benefits and potential implica-tions of this workflow are far-reaching.First, the office costs of creating as-builtdrawings are eliminated. More impor-tantly, doing design work directly insidepoint clouds can dramatically reduce therisks and consequences of “omissions.”

Models of bolt pattern on steel girder. Image courtesy Meridian Associates, Inc.

3D model of piping from scan data with true color overlay

(e.g., models of human faces), and it hasalso been used in architectural applica-tions. This type of software can be fairlycostly and typically requires advanced3D modeling competency to effectivelycreate accurate NURBS surfaces.

Extracting Volumes

Extracting volumes from point clouddata can be a highly efficient process,which also yields more accurate volumecalculations than traditional methods. To-day, volumes can be extracted betweensurfaces (including TIN meshes) or be-tween a surface and an assumed plane.The process involves creating surfacegeometry first and then extracting vol-umes and surface-to-surface deviations.This latter functionality can also be usefulfor excavation projects and for monitor-ing surface movements over time.

Extracting 3D Models

Although the need to create 3D objectmodels for civil projects is uncommon, theextraction of 3D models from point cloudsis quite well developed in advanced pointcloud software, as this is where most pointcloud processing software began. Over theyears, there has been a steady stream ofadvancements in this area that has dra-matically reduced the office time associat-ed with creating 3D object models frompoint clouds, especially for plant and facil-ity as-builts. Typically, object models gen-erated from point clouds are surface mod-els, e.g., cylinders, planes, spheres, steelshapes, other geometric primitives, andarbitrary “catalog” shapes. Tools are alsoavailable in advanced point cloud softwarefor applying standard spec tables, such asfor piping and steel shapes.

Three-dimensional object models aregenerated “piece-by-piece” from pointclouds. In some software, the user can



Many readers would be surprised athow extensively this work process is be-ing used today. It is used often in plantand facilities retrofit design projectsand has also been used in civil and ar-chitectural projects. This process is notonly used in retrofit design, but also forcomparing actual fabrication/construc-tion against design. Clients can keeppoint clouds for archive and then simplyview areas of concern themselves anddecide whether or not to extract geome-try from the data. Direct use of pointclouds is a powerful, highly cost-effec-tive work process that many predict willone day be a standard workflow.

Summary

In summary, point clouds from high-definition surveys represent a richsource of geometric information. Tools Design verification within point clouds

and work processes are available todaythat enable the extraction of traditionalpoints, 2D and 3D lines, surfaces, vol-umes, features, and models from theserich point cloud data sets. Another newdeliverable has also emerged: the regis-tered, geo-referenced point cloud itself.Point clouds can be used directly for bet-ter retrofit design and fabrication/con-struction QA, without having to first ex-tract other deliverables.

Geoff Jacobs has beenemployed by Leica Geo-systems HDS, LLC (for-merly CyraTechnologiesInc.) since 1998. He cur-rently acts as SeniorVice President, StrategicMarketing. He is also aContributing Writer for the magazine.


points, lines, surfaces, features models from point...

Documents