80 4. Static <strong>Laser</strong> Scanning<strong>and</strong> generally fulfill the constraints <strong>of</strong> orthogonalityFurthermore, a scale factor can be introduced, whichincreases the number <strong>of</strong> unknowns up to sevenThe minimum number <strong>of</strong> corresponding points definingthe transformation between the two coordinate frames is threeGenerally, these corresponding points canbe either defined by targets, which have to be modeled using the point cloud, or by using the point clouddirectlyA distinction can be made between aregistration based on artificial or natural targets <strong>and</strong> a registrationbased on overlapping point cloudsBoth methods are introduced <strong>and</strong> discussed regarding advantages <strong>and</strong>disadvantages Depending on the method, the registration describes onlya transformation between lo¬cal coordinate frames or defines a transformation between several local coordinate frames <strong>and</strong> one globalreference frameThe transformation into a global reference frame requires control points with known co¬ordinates4.2.1 Target-Based RegistrationRegistration using targets requires considerations regarding two different aspectsthe type <strong>of</strong> targets <strong>and</strong>the spatial distribution <strong>of</strong> targets Targets can be both artificial <strong>and</strong> natural In either case, the target in¬formation has to be derived based on the acquired point cloud, egcenter point, radius, normal vectorGeometrical primitives seem to be well-suited for supporting this information, such as planar objects,lplanes, spheres, cylinders Generally, the best results can be obtained by specific artificial targets Mostlaser scanner manufacturers also provide targets, wich define such tie points or control points2 Often,these tie points can be surveyed by total stations or by GPS This is helpfulsince the local coordinate frameedefined by the laser scanner can be included into other local or globalreference frames Anotheraspect isthe surveying <strong>of</strong> such targets reduces the constraint that neighbouring pointby at least three tie points or objectsclouds have to be connectedArtificial targets have to be designed according to the specifications <strong>of</strong> the laser scanner The target shouldbe dimensioned such that a sufficient number <strong>of</strong> points hit the surfaceThe material <strong>and</strong> the colour haveto have the appropriate wavelength <strong>and</strong> reflectance propertiesThe use <strong>of</strong> different reflectivity values mayease the calculation <strong>of</strong> the geometrical center <strong>of</strong> the target, e g retro-reflective targets Figure 4 5 showsthree different targets used for laser scanningThe left one is a circular planar retro-reflective target witha white inner circle <strong>and</strong> a blue outer circle, designed by HDS Leica GeosystemsIn the middle, a planarblack-white coded target<strong>of</strong> Zoller+Frohlich can be seen Theright partcontains a wooden <strong>and</strong> whitepamtedsphere mounted on a tribrach constructed byETH Zurich Allthree targets have dimensions <strong>and</strong>massesappropriate for field work Thebalance between size <strong>and</strong> mass <strong>of</strong> the targeton one side <strong>and</strong> theease <strong>of</strong> transportation on the other side are <strong>of</strong> importanceInstead <strong>of</strong> artificial targets, natural targets can also be used Often, planar objectscan be found in theenvironment, eg walls <strong>of</strong> buildings Sometimes, also cylinders or spheres are available They are goodchoices to be included in the registration process, especially if they are located in regionsin which artificialtargets can be rarely positioned <strong>and</strong>/or are inaccessibleThe accuracy provided by natural targetsare notas high as that provided by artificial targetsHowever, the use <strong>of</strong> such targetsis recommended to stabilizethe registrationBased on registration results, targets can be assessed <strong>and</strong> deactivated when theydo notsatisfy accuracy requirementsThe algorithms for calculating the mathematical point <strong>of</strong> reference for each target is either implemented inthe s<strong>of</strong>tware tools provided by the manufacturers or available in numerous s<strong>of</strong>tware tools for processinglaser scanning point clouds The accuracy achievable for the point <strong>of</strong> reference is dependenton severalaspects, e ggeometry <strong>and</strong> orientation <strong>of</strong> the target, noise within the point cloud, but can be increased upto a few millimeters2 A control point is a tie point that was surveyed <strong>and</strong> can be defined by global coordinates
4.2 Registration 81Figure 4.5: Types <strong>of</strong> artificial targets used for laser scanning: HDS Leica Geosystem circular planar target (left), ZFcoded planar target (middle), ETH Zurich sphere target (right).The distribution <strong>of</strong> targets is <strong>of</strong>ten disregarded during laser scanningfield work. The instruments are in¬vestigated <strong>and</strong> calibrated intensively to improve the accuracy <strong>of</strong> laser scanners. But in return, the accuracyis decreased through carelessness in the field caused by awkward <strong>and</strong> poor targetdistributions. The re¬sults based on the distribution worsens the accuracy significantly. In photogrammetry <strong>and</strong> analogous interrestrial laser scanning, control points have to be chosen in such a way that they• cover the full object volume to avoid extrapolation,• do not hide important parts <strong>of</strong> the object, <strong>and</strong>• are <strong>of</strong> a reasonable number.Investigations showed that for an indoor project with a dimension <strong>of</strong> 10 x 10 x 3 m3 the influence <strong>of</strong> poordistributions <strong>of</strong> control points can reach several centimeters. A reasonable number <strong>of</strong> targets<strong>of</strong> about 4 to6 for two different scan positions is recommended. The use <strong>of</strong> moretargets does not increase the accuracysignificantly unless there is a probability <strong>of</strong> unsuitable targets that should not be used for the registrationprocess.The coverage <strong>of</strong> the full object volume is essential since extrapolation should be avoided <strong>and</strong> mayresult inincreasing errors. High or linear objects, such as towers <strong>and</strong> buildings, require an appropriate distribution<strong>of</strong> the targets in height as well. Unfortunately, in most cases the attachment <strong>of</strong> artificial targets is impossible.Thus, other registration techniques should be used, e.g.natural targets or point cloud registration, <strong>and</strong>should be combined with each other.4.2.2 Point Cloud RegistrationThe point cloud registration <strong>of</strong>fers the possibility to transform one or more data sets to another withoutusing targets. The only constraint is the two data sets show areas <strong>of</strong> overlapping pointclouds. This meansboth data sets have to contain point clouds that describe the same objects. The advantages <strong>of</strong> this procedureis that no artificial or natural targets are required <strong>and</strong> a high redundancy is achieved. The disadvantagesare due to the principle <strong>of</strong> laser scanning: