Photogrammetric Procedures for Digital Terrain Model Determination

Transcription

1 Photogrammetric Procedures for Digital Terrain Model Determination Hartmut ZIEMANN and Daniel GROHMANN 1 Introduction Photogrammetric procedures for digital terrain model (DTM) data determination fall into a number of categories: Stereocompilation methods: Traditional photogrammetric capture with analytical workstations utilizing direct viewing of film imagery. Capture with digital photogrammetric workstations using either digitized or digital source imagery. Automatic collection of elevation data by digital correlation from digitized film or digital imagery. This is augmented by editing with analytical or digital compilation workstations. Hybrid approaches. Many projects involve the determination of both feature (planimetric) data and elevation data. In regard to elevations, DTM generated by these methods in the past were primarily focused on producing topographic contours. However, the recent large emphasis on digital orthophoto production has brought about the concept of elevation data sets specifically geared toward this effort; these need in general not have the same elevation accuracy und require therefore significantly less time and cost to produce. This paper will discuss the determination of elevation data. 2 Data Collection Methods 2.1 Geo-Referencing Geo-referenced images are defined as those in which three-dimensional ground co-ordinates can be mathematically mapped into a two-dimensional image space, and vice-versa. A number of components are needed for each image to be able to extract geo-referenced data from it: the interior orientation converts the two-dimensional image into a three-dimensional bundle of rays using reference marks defined as part of a camera model, and the exterior orientation provides data to uniquely position and orient each bundle in space. Depending upon image scale and required evaluation accuracies, atmospheric effect, Earth curvature or coordinate system distortions may also be corrected.

2 2 H. Ziemann and D. Grohmann No mapping of any type can begin until all acquired imagery is accurately tied to the ground reference frame. In most cases, this requires a process known as aerotriangulation; in future this process may no longer be needed as the required data will be collected with sufficient accuracy using Airborne GPS (ABGPS) and IMU data. Aerotriangulation refers to the process of measuring a number of points on overlapping images and some suitably located control points, and then determining the exterior orientation elements. This process employs the geometric sensor model and a statistical process that produces a best fit. The output of this process includes terrain coordinates for all points measured on at least two images. With aerotriangulation in mind, imagery is acquired in strips along flight lines and the strips aggregated into a block of imagery covering the project area. Imagery is overlapped within each strip, generally about 60 percent of the image width in the along-flight direction. The strips, in turn, are flown parallel to one another and generally overlapping adjoining strips by about 30 percent of the image width. Using the overlapping imagery, a number of points are measured on each, optimizing the points in multiple overlap areas: control points with known horizontal, vertical, or both horizontal and vertical position, pass points between photos in a strip and tie points between photos in overlapping strips. Most procedures focus on measuring pass points at the center, left center edge, and right center edge (comprising the classical Gruber pattern) of each image along the direction of flight; this results in nearly all points falling in tripleoverlap areas. Tie points are measured at the top and bottom edges of each image where strips overlap and thus many of these will appear on 4 to 6 images. Control points are accurately ground surveyed and either chosen as photo-identifiable points or are targeted prior to flying. They must be chosen to be in localized flat areas, free of obstructions (such as overhanging trees or nearby structures), easily identified, and sized to be accurately measured based on the image scale. Ground control point distribution throughout the block of images is generally focused on a number of points with known position and elevation along the edges of the block and a lesser number of points scattered in the center of the block. Additional control points with known elevation only are needed throughout the block. Sophisticated aerotriangulation packages simultaneously determine the ground co-ordinates of all measured points and exterior orientation elements using the method of least squares adjustment and the sensor's geometric model. All determined parameters (ground coordinates, image measurements and exterior orientation) generally include estimates of accuracy, which are used to increase the stability and rigor of the solution through the least squares process. Aerotriangulation software packages also support the so-called selfcalibration, in which deviations from known interior orientation parameters and film deformation are also determined. The use of ABGPS for exterior orientation can significantly strengthen the aerotriangulation results particularly for smaller scale projects. GPS positions effectively act as additional control points in the process. Therefore, blocks using GPS require fewer ground control points and it is not unusual for only control points to control a block of hundreds of images as long as ABGPS is employed. Incorporation of IMU information, which measures the attitude of the sensor, results in somewhat more strength in the adjustment.

3 Photogrammetric Procedures for Digital Terrain Model Determination Data Collection Systems Photogrammetric data collection has traditionally been carried out by an operator using a stereoscopic plotting instrument (stereoplotter) allowing the creation of stereo models from overlapping images. A floating mark is guided along lines of interest. In the case of contours the floating mark is maintained on the terrain surface while traversing the model methods along lines of constant elevation. In the past, such analog stereoplotters employed optical, mechanical or a combination of optical and mechanical solutions for the recreation of the imaging rays from aerial images. The invention of the analytical plotter by U. V. Helava at the National Research Council of Canada and the continuing development of computers lead from the mid-seventies to a gradual replacement of those plotters with analytical stereoplotters. These still provide direct viewing of the film images but use electronic means to recreate the imaging rays. This changeover eliminated restrictions in regard to camera type and deviations of the camera axis from verticality. A large amount of current-day stereoscopic compilation still involves the use of analytical stereoplotters. The proliferation of these systems and their proven accuracy amounts to a sure approach to generating DTMs on a day-to-day basis. The primary advantages of these types of systems are their proliferation and the large amount of published procedures and specifications surrounding their use. The main disadvantages are the manual setup of each stereomodel, continuous calibration and general requirements to operate in subdued lighting conditions and firm under-flooring, and the inability to use digitally acquired imagery. Analytical stereoplotters usually require indexing points that are derived from the aerotriangulation results (tie points, pass points, and control points) to ensure accuracy. Thus, at least some form of triangulation will be required or photo-identifiable points must be measured on the imagery to allow use of these instruments in the project flow. Software packages handling the capture of data are generally the same as those used by softcopy systems, thus ensuring combined usage within a project. Softcopy stereoplotters systems were first introduced into the commercial production environment in the early 1990s. After a period of proving accuracy and gaining operational acceptance, they began to replace analytical instruments. Some advantages of softcopy compilation workstations are the ability to use either digitized film imagery or digitally acquired imagery, no requirement to use model setup or index points, the ease of hardware and/or software upgrading to keep up with progress in technology, the relatively free sitting and head position for the user as compared to analytical systems, and the ability to be used in common office environments, though subdued lighting is still preferred or even required in some cases. The disadvantages are arguably less stereo viewing quality than direct filmviewing systems and film scanning is required for projects flown with film cameras. The greatest advantage of softcopy workstations is perhaps their extended ability to overlay graphics data in a stereoscopic environment. In the context of DTM generation, this means that all elevation points, and even contours, can be superimposed for review against the actual ground form. The ability to review existing, new, and changed data is easier if all can be selectively reviewed in the same stereoscopic display.

4 4 H. Ziemann and D. Grohmann 2.3 Data Collection Approaches A number of standard approaches are generally employed in elevation data collection, the plotting of contourlines, the capture of spot elevations and breaklines, and the collection of a point matrix. A breakline is defined as a polyline that defines a slope reversal along its extent and may include both naturally occurring and man-made features. Common examples of breaklines include ridges, valleys, edges of embankments, edges of rivers, curbs/ gutters, etc. Breaklines are augmented by a number of spot elevations that fill in areas devoid of natural or man-made breaks in the terrain surface. A number of spot elevations are collected at identifiable points to leave behind a record of points that can be checked. Spot elevations are also collected at local maxima and minima that are not represented by a contour line. Planimetric compilation can be carried out either to obtain two-dimensional (positional) or three-dimensional data. Since most planimetric features are collected at the bare earth, they can in the latter case be used to extract elevations, some breaklines (such as along a river edge or road edge) and contourlines. Additional spot elevations and breaklines may be added as needed to finalize the DTM; generally only a small amount of additional breaklines are collected but a relatively large amount of spot elevations. If positional data are not part of the overall project, or if it is scheduled at a later date, most mapping firms will use skilled operators to focus on the collection of contourlines or breaklines and spot elevations. In the hands of a skilled operator, the advantage of the latter approach is the collection of a minimal amount of points required to describe the terrain. Some mapping firms use a gridded or profile approach to the collection of elevation data, augmented as needed with breaklines and spot elevations. This is most common in very smooth areas where it is difficult to ensure that subtle changes in the terrain elevation are captured. It is also the most common approach and final format for smaller map scales. The collection of profiles was the first approach used to collect elevation data needed for the production of differentially rectified images called orthophotos. The profiles are first defined in reference to the base of a stereoscopic pair of images, and then followed automatically. A photogrammetric operator has to set the floating mark continuously onto the terrain surface. In order to facility this process, the operator can control the speed of the movement of the floating mark. Grids can be collected using two approaches, the static mode and the dynamic mode. The static grid collection mode places points by means of a grid defined in position and point spacing either in regard to a stereoscopic pair of images or to an overall coordinate system, and a photogrammetric operator adjusts the floating mark to terrain level at those predefined points. The dynamic grid collection mode is more efficient in collecting only points needed to actually represent the terrain surface. An initial static grid is measured and continuously refined by reducing the grid spacing using a factor 2 as long as the elevation value differences determined for the four corners of a grid cell exceed a defined tolerance. A minimum grid spacing is used to stop the process in rough areas. Automatic collection of DTMs by means of digital image correlation has been a topic of research and implementation for over 40 years. With the advent of more powerful desktop computers over the last decade, more robust and intensive algorithms have been implemented. Automatically deriving elevations by digital correlation fundamentally amounts to

5 Photogrammetric Procedures for Digital Terrain Model Determination 5 directly comparing patches of pixels on conjugate images or indirectly comparing information derived from the digital images. Direct comparison techniques were the first to be researched, with increasing work done in comparing derived information, such as features defined by edges in the images. Edge/feature collection techniques tend to be superior in large scale projects, with area-based techniques more common for smaller scale projects. One of the keys to successfully employing automated approaches in day-to-day production is the understanding of the strengths and weaknesses of the technology as a whole. General strengths include the ability to generate a very dense array of elevation points in short periods of time in an unattended mode and a generally good capture of the overall surface of a ground scene especially at smaller map scales. General weaknesses include the need to edit large amounts of data that do not necessarily reflect what would be captured interactively by a human stereo-operator, disregard to breaklines and the difficulty in effectively deriving data for larger product scales. A hybrid approach to data collection may be advantageous and for smaller map scales very effective. Digital correlation can be used to quickly develop a surface over large areas, leaving final editing to softcopy workstations. This hybrid approach supports interim or expedited delivery of digital orthophotos, with the softcopy compilation used to develop breakline information required for topographic contours. 3 Data Collection Issues The foregoing discussion leaves several important issues untouched; six of these will now be introduced. 3.1 Introduction of Digital Aerial Cameras Two digital aerial cameras were introduced into the market recently, the ADS40 by Leica Geosystems and the DMC by Z/I_Imaging. The ADS40 uses linear arrays collecting image data with forward motion (pushbroom system) and acquires stereo images by at the same time looking forward, down and aft. The DMC uses four area arrays to collect partial images later combined into a single rectangular image. The introduction of digital aerial cameras results in significant operational changes. First indications are that accuracies valid for operator-derived data from aerial photographs not longer apply. The authors are involved in the organization of a first operational test aiming at deriving such accuracy reference data. 3.2 Extraction of Features The emphasis in this discussion has been on elevations. Topographic features can be derived by point measurement and line following providing vector data sets. Another approach consists in the derivation of differentially rectified images (orthophotos) which are often combined into mosaics covering larger areas. This approach requires exterior orientation and elevation data. The process is often used when a map is required in a short period of time, especially since the process is highly automated in particular when digital image data and softcopy plotters are used.

6 6 H. Ziemann and D. Grohmann 3.3 Elevation Data Accuracies An important aspect in the generation of a three dimensional DTMs is achievable accuracy. Experiences summarized in Kraus may serve as guidelines for different methods carries out by an operator: contourlines grids profiles 3.4 Image Matching Procedures σ Z = ± 0,20-0,25%o c m I σ Z = ± 0,10-0,15%o c m I σ Z is dependent upon terrain slope and scan speed In photogrammetry matching can be defined as the establishment of correspondence between various data sets, in particular between images. Examples of image matching include interior orientation (the image of a fiducial is matched with a two-dimensional model of the fiducial), relative orientation and point transfer in aerial triangulation (parts of one image are matched with parts of another image in order to generate tie points), absolute orientation (parts of an image are matched with a suitable image of a ground control point) and generation of DTMs (parts of one image are matched with parts of another image in order to generate three-dimensional object points). Image matching procedures can be divided into groups using different approaches, one being based on the two categories of matching primitives, area-based matching (ABM) and feature-based matching (FBM). ABM uses windows composed of grey values which can be extracted fast, and the actual matching process is rather straightforward and in the context of this paper of the greater interest. FBM uses local features like points, edge elements or small region which are first extracted in each image individually. Hierarchical methods are used in many matching algorithms going from coarse to fine, and the results achieved on one resolution level are used as approximations for the next finer level. It is therefore desirable to define images in a variety of resolutions, leading to the socalled image pyramids. 3.5 Use of Satellite Images A number of commercial remote sensing satellite systems are now producing high quality and high dynamic range images. Currently, the highest resolution is 0.6 meters/pixel. Satellite systems vary in their means of acquiring stereo images. Different-pass collection systems are generally based on linear arrays collecting with the vehicle's forward motion, and stereo is generally acquired by rolling the spacecraft axis outboard (normal to the direction of flight) on successive passes. These systems are prone to localized weather conditions between imaging events, which can be as high as 10 days. Same-pass systems acquire by pitching the spacecraft axis forward or aft preceding or succeeding the target area, respectively. These are also generally pushbroom scanners and are the most optimal systems to use in mapping. Satellite systems vary heavily in their design and operation, and satellite geometric sensor models are far more complex than frame cameras or even digital airborne systems. Satellite imagery is very conducive to automated approaches involving elevation extraction, and its availability has spawned a growing demand for the creation of

7 Photogrammetric Procedures for Digital Terrain Model Determination 7 DTMs for visualization databases for fly-throughs and perspective scene generation, applications requiring less accuracy. 3.6 Non-Image Procedures A LIDAR (Light Detection And Ranging) system is a complex package of electronic components finding increasing use for the determination of DTMs. It includes a LASER (Light Amplification by Stimulated Emission of Radiation) for the measurement of distances from a platform, an INU (Inertial Measurement Unit) to determine the orientation of the laser axis in space and a GPS (Global Positioning System) to measure the position of the sensor in space. Lasers operating within the infrared part of the electronic spectrum are used over land. They emit pulses which are reflected and returned. Sensors capable of recording multiple returns are available. A first return usually provides information on the elevation of the canopy, a last return the elevation of the ground. At present accuracies in the order of σ Z = ± 0,15 m can be achieved. SAR (Synthetic Aperture Radar) takes advantage of the motion of a platform to synthesize a large antenna to achieve fine along-track resolution. The RADAR (RAdio Detection And Ranging) points in a direction perpendicular to the flight path and measures ranges. By augmenting a SAR system with a second spatially separated receiving antenna it is possible to extract topographical information; such a system is called IFSAR (InterFerometric Synthetic Aperture Radar). They achieve at present accuracies in the order of σ Z = ± 0,5m. 4 Conclusion A review is given of photogrammetric procedures to derive digital terrain models with an emphasis on elevation data. New technological developments are indicated which will cause significant changes to traditional procedures. 5 References Heipke, C.(1996): Overview of Image Matching Techniques. OEEPE Official Publication No. 33: Maune, D.F., (ed) (2001): Digital Elevation Model Technologies and Applications: The DEM Users Manual. American Society for Photogrammetry and Remote Sensing, Bethesda, MA 20814, USA, in particular Chapter 5 (Molander, C.W.: Photogrammetry)