System failure or instabilities shall be mentioned also: e.g. change of the set of available GPS satellites during a strip might cause some displacement; however, IMU data helps to bridge such critical gaps. Last, but not least, the missing rigorous supervision of the whole measuring process has to be mentioned. Ignoring this and accepting such “raw“ data, merging the strip point-clouds yields rather rough DEMs even in flat areas! Instead of the minimal solution cited above we propose the subsequent alternative: Use ground-reference. We are conscious of the fact that ground-reference-points can’t be “identified” in the point clouds of the laser-scanner strips. But we can use “features” – e.g. plane patches – which are supported by traditional ground reference-points (see 2.1). For planimetric fitting, roofs of buildings and/or prominent fault lines in the terrain are suitable, for mere height fitting, horizontal areas free of vegetation are recommended. In photogrammetric terminology, we call those reference-points usually control-points. Monitoring a many of plane- and height-discrepancies in the common areas of overlapping laser-scanner strips and, therefrom, improvement of GPS-positioning and IMU-attitude data. Mathematically formulated with correction polynomials (of quite low degree) for the recorded orientation elements as function of time: one strip – one polynomial. This procedure preserves the high neighbouring precision of both system components and copes with drifting phenomena. Adjustment has to be done simultaneously for all strips of a block – using the positions of corresponding points (features) in the overlapping areas as observations minimising their residuals. A statistically better approach lies in using original observations (Kraus, 1997a): the polar co-ordinates recorded by the laser-scanner; given position and attitude of the scanner, the Cartesian ground co-ordinates are (simple) functions of those recorded (ν, χ ,ρ)values, i.e. scan-angle ν , fore-sight angle χ and distance ρ . The above outline of a technique to improve the geometric quality of laser-scanner data should give an idea how to overcome gaps between strip surfaces. Unfortunately, the proposed method requires access to the original data of the laser-scanner: GPS, IMU, and Polar data as function of time. The laser-scanner companies want to provide 3D-data for the end-user – so, they want to provide “DTMs” (i.e. grids) resp. point clouds in the national ground-survey co-ordinate system, only (“user-friendly”). But this “end-product” is prone to having bias and is too late in the processing-chain for elementary repair. Nevertheless, we have to stress the fact that our criticism is valid only when exploiting the full potential of laser-scanner data: we want to get the few-cm-precision of the laser-scanner also as accuracy of the end product. Provisorily (Kager, Kraus, 2001), we tried an approach based on raw 3D-XYZ-data given in the national co-ordinate system strip by strip. Instead of correcting flight path (dGPS) and attitude data (IMU), we tried to compensate for the apparent XYZ-deformations by correction polynomials for individual strips of ground points. This procedure has the disadvantage that it copes merely with phenomena and does not assess the true problem. But it had the advantage that the necessary data is easily available to end-users. We got an improvement, but were not really satisfied. Nowadays, we aim at a strict, highly automatable procedure minimizing 3D-gaps. Since trajectory data is a prerequisite of this method, the polar co-ordinates may also be re-calculated from the ground co-ordinates provided their time-stamps are given. Anyway, such data is deliverable by the companies. Before going into adjustment details we have to discuss the determination of strip-tying features. Determination of Strip-tying Features The principle of strip-tying by features is shown in figure 1. As we are not able to associate homologous points in the point-clouds created by LIDAR, we have to recourse to simple geometric features like planes which can be derived from suitable regions of LIDAR-points.  Figure 1: Principle of block adjustment with three concurrent laser-scanner strips using planar features At some suitably chosen ground position XY, a plane can be interpolated into every point-cloud of overlapping strips. Since the available orientation of the raw strips is relatively good, we can expect that the homologous features will also overlap. It should be mentioned here that the term “feature” also includes lines (straight or curved). But this aspect should not be followed here in detail, since a line can be conceived as intersection of planes (surfaces) and handled by these means. "homologous points" vs. "homologous planes" A point has three co-ordinates: geometrically spoken, the three co-ordinate planes intersect in one point. Alike, any three intersecting planes yield a point. So, three neighboured homologous planes together are equivalent to one homologous point. Orthogonality would be optimal. Homologous plane features consist of regions of about 5 to 20m extension; for shortness, we call such a plane-feature “patch”. See Figure 2. The above deliberations also hold true for control-points. We have to replace control-points by control-features: We determine geodetically four supporting points for one patch plane. See also Figure 2. The fourth (superfluous) point serves for checking and over-determination purposes.  Figure 2: Examples of three tying-patches providing different expositions in the patch-set. The Patch-Finding Mission We use chronological data of the LIDAR-strips, since this data-structure preserves topology to a high degree whereas a point-cloud has to be considered topologically unstructured. The usual procedure on giving a point-cloud again a topology is triangulation yielding a TIN. But this is time consuming and in the XY-domain sometimes wrong (e.g. a point on the wall might appear inside the eaves of a house). Since we want to use original data, i.e. unfiltered data, we don't want to use a regular (desirable), but interpolated (regrettable), grid. Proposition: A topology in the domain of time and scan-angle as seen from the trajectory is free of loops. (There is one easy to be handled exception: "over-scanning" caused by rare pitch-change making the scan-line move even reversely.) We search patch-candidates in this topology. A patch-candidate is now a (tilted) plane matching some criteria: it is above the surrounding (case roof) is planar within some tolerance (e.g. standard deviation 0.04m) has minimal steepness (case roof) has not too many outliers (ignore chimneys, dormers, etc.) has minimal count of supporting points (not too small). etc. Adjustment with data-snooping of a general plane is used to determine patch-candidates. Every strip generates a list of patch-candidates where a patch is represented by its unique patch-identifier reference-point (chosen centre of the region) normal-vector incl. accuracy shift along the normal incl. accuracy four anchor-points circumscribing the region bearing the attributes: time , polar co-ordinates (scan-angle  , fore-sight  , distance  to the adjusting plane; they represent the many of original polar points and will be used in adjustment as observations etc.This first run through the data gives for every strip a list of such patch-candidates. In a second run, for every overlapping strip these (accordingly sorted) lists are used as seeds for determining the respective homologous patch-candidate. So, an original patch-candidate may get no, one, or more partners (e.g. from cross-strips). Any strip produces now a second list of “homologous" patch-candidates. The structure is the same as above. It is noteworthy that all homologous patch-candidates bearing the same patch identifier are of equal rights concerning the least squares of adjustment theory; no correlations between the observations of different strips are introduced. This second run has an additional criterion in determining the plane: compatibility of normal vectors. Having these two sets of lists of normalized patches, they serve as input for the adjustment programme. Patches which have no partner are cancelled. |