Read MountainShadows.indd text version


Introduction: Natural shadows whether viewed or captured in photographic images are a normal part of world scenes, mountainous or not. While they represent a major element of landscape perception, they can cause some interference in cartographic representations. In hill-shading work, especially in the high mountains, shadows are often subdued or omitted to allow data to be clearly drawn in normally shadowed areas. In Imfeld's work on Mount

Fig 1: Mount Rainier sunbreak Blanc shown at the left, he used light shadows on the glacier, cast from a northwest light source. In Patterson's work for the US National Park Service, he chose to omit the shadows normally cast by the high walls of the caldera at Crater Lake in

Fig 2: Xaver Imfeld, "La Chaine du Mont Blanc" (section), 1896

Oregon. When a perspective view is chosen, the panoramists such as Berann and urban illustration firms such as Bollman Bildkarten Verlag make sparing use of shadows or omit them altogether for the information that they obliterate. In stereo aerial photographic missions for mapping, flight times are chosen where possible to minimize the shadows that interfere with stereo-plotting, however in missions flown for interpretation, some degree of shadow is desirable to enhance the identification and interpretation of natural and cultural scene elements. While shadows can often represent limitations to the application of aerial photographs, on occasion they can be useful for several different applications. This presentation is a report on two lines of Fig 3: Thomas Patterson, "Crater Lake National experiments with the application of aerial photoPark" (section), 2002 Page 1

graphic shadows for different types of measurement. The first is evaluating the accuracy of US Geological Survey Digital Elevation Models (DEMs) and the second is an exploration of using shadows as a method of geo-referencing aerial photographs. Hawai`i's tropical mountain terrain is the result of both large landslides and severe erosion in a volcanic landscape. In the windward areas, very steep slopes and narrow valleys can cast shadows across the width of valleys and reach offshore areas. At the very top of a feature, the slope is gentle, but the angle of slope changes to very steep, very quickly, given the erosional pattern in this volcanic landscape. These general patterns are affected by the age of the individual islands, with the oldest being much more greatly dissected than the younger easternmost islands. Fig 4: Na Pali Coast, Kaua`i, Hawai`i. Characteristic of all of the higher islands is a large density of similar valleys and drainage networks which often makes identification of a specific valley difficult, as attested to my many lost hikers. With its steep valley walls, narrow valley floors and dense vegetation, shadows are a very important part of the Hawaiian visual environment. Shadows and perception: Cartographers, photo interpreters and relief artists know the importance of shadow orientation when interpreting planimetric images of terrain. Aerial photo-interpreters always turn photographs so that shadows fall toward the viewer to avoid the pseudoscopic effect. In this effect, the relative elevation of terrain features appear to invert when the shadows fall away from the viewer. Similarly Fig 5: Pseudo-scopic effect, relief inversion by changing shadow orientation. the conventional "northwest" lighting direction of shaded relief artists produces the same effect of shading falling toward the viewer. In much of the northern hemisphere, the sun is never in this "northwest" position, but the goal of terrain realism is over-ridden by the need to defeat the pseudo-scopic effect in shaded maps. However, when the image moves from the vertical or planimetric position to the oblique or perspective, the Page 2

pseudo-scopic effect seems to lose its importance very quickly. In this case, the vertical dimension of the relief overrides the shadow effect in almost all situations. There is, however, an element of shadowing that reaches far beyond the degree of realism that they contribute to a scene, real or rendered. Shadows contain significant detail about the terrain features that cast them and features that receive them. While there is a large amount of intuitive knowledge of the meaning and interpretation of shadows, to evaluate them quantitatively is much more difficult. There are many digital tools that can create shadows but very few tools that can describe them by a statistical or a mathematical function that would allow classification or comparison. One of the tools that is missing is a robust shape index. If we had such a tool, many of the procedures that follow could be automated and made much more generally useful. Terrain rendering software: In the last two decades, there has been an explosion of the so-called 3-D object creation and rendering packages. Starting from the long-standing engineering design programs created by Autodesk, many other types of terrain and 3-D object rendering software have become available on platforms ranging from mainframes to the smallest personal computers. The different modelers have developed in several fairly well-defined niches ranging from the high-realism complex packages used in film animation and advertising to the data display software designed for scientific visualization. Many of the programs can introduce terrain and vegetation into their scenes, but only a few are designed specifically for terrain visualization, or easily amenable to it at the variety of scales and perspectives useful to cartographers and earth scientists. In general, the available terrain packages fall into three different categories by how they deal with slope, aspect shading and shadows. Ray-Tracers: The category of renderers that use a ray-tracing algorithm (Bryce, Vue d`Esprit, Natural Scene Designer, POV-Ray and others) shade and cast shadows by rendering a view in RAM and then adjusting the colors and brightness of each rendered image by tracing a "ray" from the camera or viewpoint through each pixel in the view. `Behind' the view plane, each ray is reflected, refracted, or transmitted by stages back to the light source. Each pixel is adjusted according to the nature of the objects it encounters. In a detailed and high resolution scene, the ray tracing process is computationally time-intensive but produces very high-quality rendering and automatically shades slope and develops shadows according to the position, color and nature of the light sources. In most of these renderers, the light is moved manually or by specifying azimuth and elevation angle from the object or scene.

Fig 6: Hypothetical scenes rendered by ray-tracing software (Bryce 5) Fig 7: Water reflections in a ray-tracer (Bryce 5) Page 3

The darkness of the shadow is controlled by the algorithm and how it is modified by the amount of ambient light defined in the scene. Note the softness of the shadow at the peak of the cone's tip shadow Figure 6). One very convincing aspect of ray-tracers are the way that complex reflections are very realistically rendered. Slope and Aspect Shaders: The terrain render programs that adjust the rendered tone of a slope according to slope and angle from the illumination do not cast shadows. As the sun reaches higher angles in the sky the light falling on plane surfaces increase and the slopes that are normal to the sun exhibit the maximum brightness. The slopes that are steepest and aligned away from the sun have the darkest tones.

Fig 8: Shading by slope and aspect of hypothetical terrain with a west light source (Surfer 8) Fig 9: Terrain shading by slope and aspect (Surfer 8) The shaders (Surfer, MicroDEM ArcView and others) generally are very fast renderers because the algorithm for the shading is very simple and does not involve the multiple stages of computation that the other types of renderers require. Shadow Mappers: A class of programs that compute shadow maps and apply them to the terrain in addition to the shading functions are the Shadow Mappers (World Construction Set, Visual Design Studio and others). They fall in between the Shaders and the Ray-Tracers in time required but can also be used in real time applications. There are many different types of Shadow Map algorithms available. Despite the differences in methods of processing data, the application of the shadows are quite similar to the Ray-Tracers, however, the slopes facing the illumination can be somewhat different in their tone application. In ray-tracers, shading, shadows, and reflec-

Fig 10: Shading and cast shadows by a shadow mapper (Visual Nature Studio 2)

Fig 11: Terrain render by shading and shadow mapping (Visual Nature Studio 2) Page 4

tions are combined in the same function. In shadow mappers, the three functions are separated into different algorithms and can be independently controlled to fine tune the sum of the different effects. Shadows cast and simulated: While working with a topographic mapping team in the valley Nualolo Kai on the island of Kaua`i in Hawai`i, it became obvious that the wide valley with very steep walls had a very interesting configuration for analyzing shadows. The steep walls are very good for making comparisons between real cast shadows from aerial photographs and ground observation to those made from DEMs in Shadow Mapper and Ray-Tracer software. In Nualolo Kai, the ability to predict shadow location was useful to field work planning because the mapping team could only spend three or four working days in the valley twice a year so the scheduling of specific tasks Fig 12: Nualolo Kai from Alapi`i Point. according to the high heat of working in certain locations made the limited work time more efficient. In addition, the valley was very good for evaluation of shadows because low vegetation made it easy to track the movement of real shadows across the slopes at all times of the day. The quality of the shadows predicted by software are a function of the shadowing algorithm and the quality of the Digital Elevation Model used as input to the software. Real and photographic shadows are also affected by the vegetation on the terrain while predicted shadows can be controlled by parameters that can be adjusted in the software. Ray-tracing and shadow mapper software were evaluated for exploratory work on shadow prediction in Nualolo Kai. I am most familiar with the ray-tracer Bryce 5 from Corel and the shadow mapper Visual Nature Studio 2 by 3DNature, so these were used for the comparison. A crucial element in the design of the two softwares is control of the lighting. At the present time, Bryce is not geo-referenced. This characteristic means that sun controls are based on a location determined from the scene itself. Light angle can be interactively controlled by the illuminated black ball on the right side of main control panel below. As this control is rotated, one can see the effect in the small "nano-render" window on the left side of panel. It is very difficult to control the illumination location precisely with this control. In Bryce's Sun Controls section of the Sky Lab pane, the azimuth and altitude angle of the sun can be entered in decimal values. The Sun Control allows much more control of shadows than the main menu, however the user must toggle in and out of Fig 13: Bryce's Sky & Fog edit panel. this screen and do a Page 5

preview render to see the effect of the change. In addition, to simulate a specific time of illumination, computation of the location of the sun in the sky must be done to convert location time into azimuth and altitude. By contrast, Visual Nature Studio (VNS) is geo-referenced and includes an internal calculator to position the sun by local solar time related to the latitude and longitude of the DEM. A program algorithm uses the latitude and longitude of the DEM to compute the azimuth and altitude of the sun and translates that to the Latitude and Longitude of the overhead sun position. Fig 14: Bryce's Sun Control pane. VNS uses a real-time Open Graphics Library (OGL) display system for its working panels while Bryce uses a low resolution wireframe display for speed of refresh. This OGL system displays shadows but the resolution of the display only provides the detail necessary for a fast inspection of the shadow effects. Like Bryce, a preview render in VNS is necessary for a detailed inspection of the shadows. VNS' ability to position the sun by local latitude and longitude, made it the choice for the Nualolo Kai experiments. Fig 15: VNS' Light Position by Time dialog box. DEM shadows in Nualolo Kai: The geographic characteristics of Nualolo Kai, noted before, appeared very useful in evaluation of DEM accuracy by comparison of real and rendered shadows. The geographic location of the site needs to

related to local time that is used by the software to determine the location of the sun. The goal of these preliminary experiments was to determine how closely VNS can mimic photographic shadows. Large scale aerial Fig 17: VNS Preview render. photographs are quite Fig 16: VNS OGL preview. rare in this part of Kaua`i. A frame containing Nualolo Kai was acquired from a 2000 coastal survey by the National Oceanographic and Atmospheric Administration (NOAA). The frame is one of a series used by NOAA to monitor coastal change in the Hawaiian Islands. The frames are taken periodically from high altitude when the weather permits. The frame used for these experiments was taken April 20, 2000 at 2:28:34 GMT. This late afternoon photograph contains strongly defined and deep shadows, ideal for shadow prediction matches and the floor of the small valley in the sun. Page 6

Converting Greenwich Mean Time (GMT) to local solar time is necessary to properly configure the software. The process of converting first adjusts GMT to Hawaii Standard Time (HST) and then refines HST to West Kaua`i local solar time. The conversion process uses the difference in longitude between the central meridian of HST and Nualolo Kai. To get the local time within minutes, the Equation of Time (EOT) is applied, Fig 18: North oriented and cropped section of NOAA #1418 a function that relates the earth's date to the planet's inclination and ellipticity of the earth's orbit which creates a positive or negative deviation from mean time depending upon the time of the year. To keep all of the Hawaiian Archipelago in the same time zone, the zone is wider than most other approximately 15 degree zones around the world. Hawaii Standard Time shares a central meridian with Alaska Standard Time. The HST central meridian of 150 degrees west, well to the east of all of the Islands. For this reason, the sun is at its zenith on O`ahu (157 degrees west) at approximately 12:30 pm HST. Since western Kaua`i is considerably further to west, the sun is overhead about 39 minutes after HST noon. Including the 1 minute 9 second advance from the Equation of Time, local solar time in Nualolo Kai is approximately 37.5 minutes behind HST. This longitude and EOT converts the 2:28 GMT time of the photo acquisition to 17:51 (3:51 pm) Nualolo Kai local solar time. The difference in the EOT can be up to 16.4 minutes fast in November or 14.2 minutes slow in February. A series of Nualolo Kai images from the 10 meter US Geological Survey DEMs were prepared with shadows to compare to the NOAA image. The principle point of the NOAA photograph is near the center of the cropped frame in Figure 18, so the shadow cast from the eastern wall of the valley should provide the best area of correspondence to the VNS rendering. Matching Photographic and Rendered Shadows: First visual inspection of the shadow matching set for 3:51 pm shows some deviations in several locations, especially at different elevations, but in general it is quite a reasonable overall fit. The southern highland areas have to be omitted from the evaluation because of the shadows cast by the clouds in the photographic image. However there are obviously great limitations to visual inspection of images adjacent to each other. To help create a data set that was easier to compare, the shadows from the aerial image needed to be converted to a vector pattern that could much more easily be referPage 7

enced to the VNS shadow pattern by overlaying the photographic shadows. The vector conversion was done by using a PhotoShop and AutoCad. The aerial photographic image was filtered in PhotoShop's "contour trace" filter to develop shadow edge lines from the shadowed image at a level that would outline the dark black shadows (Figure 20). This image was then blurred by a PhotoShop filter to increase the width of the lines to make them easier for noise editing and vectorizing. (Figure 21). The image was then converted to a binary bitmap for vectorizing (Figure 22). The goal of this process was to develop a set

Fig 19: 3:51 pm Nualolo Kai shadow comparison from image and VNS shadow mapper

Fig 20: Section of PhotoShop contour trace of NOAA 1418 of shadow edges that could be overlain on the predicted shadows to check for fit between the two data sets and to try to identify differences that might arise from DEM accuracy and photographic characteris- Fig 21: PhotoShop blurred contour trace of 1418 tics such as tilt, scale distortion and radial displacement. After the bitmap was created, it was imported into Autodesk's Raster Design program for conversion to vectors. It was not important that the image be perfectly vectorized to reflect the complete shadow edge set. Most important was the ability to note the difference in the location of distinctive contours Fig 22: Blurred shadow contour trace bitmap of the shadow on the two data sets. Once the rough Page 8

vectorizing was completed by the polyline follower in Raster Design, the VNS shadow image at was entered into the AutoCad drawing. In this form, the average visual analysis is not as nearly as convincing as in the prior inspection. It is obvious that the considerable radial displacement in the image is moving the shadows significantly

Fig 23: Overlay of NOAA 1418 shadow edge vectors on VNS shadowed image. Shadow vectors were fit to the DEM image from coastline shadows beginning from the coast nearest the image principle point of the aerial photograph from their ground position. As expected, a fit at the 1200 foot level (not shown) makes the upland shadow fit much better but at the expense of the coastal shadows. Digital Elevation Model Accuracy: It was expected that a comparison of the shadows cast by DEM modelers, and as represented on an aerial photograph, would offer insights into DEM accuracy. Since USGS DEMs are made from 1:24,000 topographic map contours with a 40 foot interval in this area, the DEM is dependent on the accuracy of contours. It is very difficult to develop a general metric of DEM accuracy. It is acknowledged by the US Geological Survey that contours have variable accuracy in different types of terrain and vegetation cover. National Map Accuracy Standards (NMAS) specify that 90% of tested points in the contour network shall fall within one half the contour interval of their true location. However, this specification can be applied in either a vertical or horizontal component. Unfortunately, only a very few points are tested for NMAS compliance and usually at locations Page 9

that afford easy surveyor access. Because of the difficulty in stereoplotting in very steep, complex terrain, contours in this area are assumed to be less accurate than NMAS specifications. The spot heights on topographic maps can provide clues to the general accuracy but in the case of rugged mountainous tropical terrain, their locations are difficult to locate on derivative DEMs. In only a few locations along the Napali Coast, are there locations that allow direct comparison between spot heights and DEM points that have any degree of confidence. There have been several different methods and software used by the USGS in preparing DEMs that have different data production characteristics. The first work in DEM preparation at the US Geological Survey was in support of their orthophotograph production activities. These DEMs, termed level 1, were usually made at a 30 meter resolution with rather loose vertical accuracy specifications. They were adequate for producing orthophotographs but exhibited a great deal of inaccuracy and systematic noise and localized errors due to mis-coding of the contour input. Because of the accuracy characteristics of Level 1 DEMs, they are not adequate for terrain representation except at the smallest scale. US Geological Survey Level 2 DEMs are designed for terrain visualization. These DEMs are processed often at a higher resolution with much more attention to the elimination of systematic noise factors. Most of the Level 2 DEMs are prepared at a resolution of 10 meters and have more stringent accuracy characteristics. Level 1 DEMs have a desired RMSE accuracy of 7 meters, however a maximum error of 15 meters is allowed. Each DEM is tested by evaluation of 20 interior points and 8 edge points. Level 2 DEMs maximum error is one half the contour interval with the same check density. The Nualolo Kai experiments were conducted with 10 meter Level 2 DEMs. DEM Shadows and Real Shadows: Although the shadow correspondence between aerial photographs and DEMs is recognizable and their differences suggest DEM accuracy limitations, the effect of the characteristic radial displacement of perspective aerial photographs overrides any detailed analysis without a great deal of processing and computation. To work with recorded shadows, aerial photographs need to be processed into orthophotographs. In this area of Kaua`i, the largest scale orthophotograph is at a scale of 1:24,000 with a shadow pattern that covers most of the valley. Therefore, it is not possible to use this image for evaluation of shadow pattern. Cost precludes flying imagery at an appropriate scale and time to prepare a custom orthophotograph. The only practical way to determine if the shadows cast by DEMs are useful for DEM accuracy evaluation is to take photographs in the valley at known local times. On a recent trip into Nualolo Kai, a number of photographs were taken in different shadow situations to determine if a terrestrial view of cast shadows would provide more insight into the evaluation of DEMs. The photograph to the right was the best of the series taken over three days for this evaluation. By duplicating the camera location on the Fig 24: Kamaile at 8:33 HST (7:54 local time), phoDEM as closely as possible, the following two shadow rendered frames visually seem to bracket tograph by Allen Hoof, September 4, 2004. Page 10

the actual photograph. The trees were not inserted into the renderings which confuses the interpretation on the right side of the rendered frames, but there are immediate apparent problems. The first is the generality of the representation of the terrain by the DEM generated from the contours of the

Fig 25: VNS shadows at 8:10 local time

Fig 26: VNS shadows at 8:20 local time

valley. The US Geological Survey notes that contours are less accurate in steep terrain, but the difference between the photograph and the view seems quite large indeed. The generalization is especially noticeable in the complex ridges at A and the profile of Kamaile at B. The local time difference between the photograph and the renders is estimated to be approximately 16 minutes. Since the top of Kamaile is not tree covered to affect the cast shadow, the difference in time must be due to one or a combination of the three following factors. (1) a difference in the DEM and real terrain elevation due to cartographic generalization; (2) a difference in the location of the horizontal location of the slope between DEM and terrain: or (3) a computational problem with the shadow mapper algorithm. A strongly shadowed scene later in the same morning provided another opportunity for inspecting the shadows received on a smoother surface. The three critical areas for comparison is the beach area (A), the very steep pali (cliff) at (B), and the talus slope at (C). The shadow on the beach, just after 9:24 local time (Figure 28), gives a quite realistic view, however the pali is not shadowed and the talus slope is still in a partial shadow. At 9:44 local time (Figure 29), the talus slope is in sun as well as a part of the beach which was not in sun in the photograph. An exact match will never be possible due to the inherent inaccuracies of the DEMs and their source topographic maps. Fig 28: VNS shadows at 9:24 local time Fig 27: Beach scene 9:44 local time Page 11

ShadowPrediction and DEM evaluation: The differential fit of the VNS shadows to an actual scene photograph appears to be a function of the DEM preparation accuracy, which in this series, is based upon the contours on the US Geological survey topographic maps. Good results have been reported predicting lineof-sight satellite contact by shadow casting in VNS. However, the local, large-scale prediction of a detailed shadow pattern will have to await more accurate DEMs, generated directly from aerial photographs, or other sensors, without the intermediate use generalized contours Fig 29: VNS shadows at 9:44 local time designed for smaller-scale applications. The Use of Shadows to Geo-reference Historical Aerial Photographs: Recently, a nearly complete set of approximately 1:12,000 stereo aerial photographs of the island of O`ahu taken in 1927-29 was found, after being missing for a number of years. Most of the missing blocks of photographs are of the urban areas of the island which would be fairly easy to index and geo-reference even with the time elapsed. Many of the urban photographs could be converted to orthophotographs with existing control and cultural reference points. A large proportion of the remaining set of photographs were taken in the high mountains at a relatively early hour. This timing of the flights, while best to avoid clouds and turbulence in the tropics, cast rather large shadows in many valleys on many of the flight lines. While working with the prediction of cast shadows in Nualolo Kai, I had started to scan, index and geo-referFig 30: Deeply shadowed imagery characence the flight lines of the O`ahu set. When working teristic of the 1927 series with a 1927 set of large scale topographic maps, it was very difficult to fit the images to the terrain for a simple index. It seemed reasonable that if O`ahu could be rendered in VNS with a set of shadows at as close to the flight time of a flight line as possible, the shadows could add a very important additional clue for indexing and georeferencing. To test this proposition, a flight line and photograph was chosen in the Waianae Mountains that had relatively flat valleys separated by steep and complex ridge structures.

Fig 31: Test image OG-82 in the Waianae Mountains of O`ahu

Page 12

A Time Matrix: The first step in the experiment was to determine as closely possible, the local time that the image was taken so that a geo-referencing base could be developed in VNS. Because of the shadows, the image had to have been taken either early in the morning in the sum-

Fig 32: First stage of shadow / time matching mer or in the morning later in the fall or spring. Since the approximate location of the frame was known from features on other frames in the line, a general area was set up in VNS using 10 meter DEMs of the area. An "intersection" plot (Figure 32) was made by estimating the time of the frame and then rendering each month at this time. After rendering the month sequence, a full matrix of likely shadow renders was prepared and Fig 33: One hour interval shadow matrix then the most likely frames for a shadow match were selected by visual inspection. The purpose of developing the best shadow match is not to determine the time or date of the exact exposure but to create a shadow base for indexing aerial photographs. The shadows cast in the fall and spring will be nearly identical in the tropics. One would have to rely on other vegetative clues to determine the exact date and time of the image exposure. The importance of determining a shadow match is to refine a time for defining a VNS light position for the shadow render. When a sample VNS render closely matches the model aerial photograph, a regional shadow render can be executed that should allow an accurate positioning of photographs along the flight line thus developing the georeferencing coordinates of the corners of the photograph. When working with indexing historic aerial photographs, the slow flying aircraft were on a flight line long enough that a noticable shadow change would occur within a photographic series. Page 13

Refining the Fit between Predicted and Imaged Shadows: The arbitrarily chosen test date for the time matrix generation was not at mid-month, therefore further refinement of the fit was useful. The best shadow fit in the matrix was chosen as 10 am local time, October 24. To refine this estimate, 15 minute deviations from 10 am were rendered. After preparing this set, a 9:45 am local time frame was chosen as the best match. Renders were then made on each fifth day on both sides of October 24 to better refine the match.

Fig 34: Fine tuning the time match To discern the small differences at this level of change, each of the VNS rendered frames were converted by the Edge Detection Filter in PhotoShop. The shadow edges become quite very well defined in contrast to the fractal noise of the DEM with this filter. The choice made as the best fit was November 3 at 9:45 am local time. It would be possible to further refine the best fit parameters, however with the lack of a tool to quantify the areal fit of two complex polygon distributions, it was felt that there was no real benefit in attempting further refinement. Fig 35: Fine tuning the date match Spatial Fitting Tools: Visual Nature Studio 2.5: The basic tool for fitting shadow maps is the shadow casting function of Visual Nature

Studio. All of the previous renders were produced with VNS on an area larger than the sample test frame OG-82. To provide a base shadow rendering for the flight line that contained OG-82, a smaller scale render was done at the best fit time and date. In addition to a time-tuned shadow render, VNS offers the ability to render scenes that are geo-referenced to a range of coordinate systems. This feature allows the scene to be input to into other GIS and geo-referenced software such as those produced by Fig 36: VNS region render at 9:45 am local shadows on November 3 ESRI, Autodesk, and others. Page 14

PhotoShop CS: Adobe's PhotoShop has several features that makes it very useful for preparing shadow images for fitting and geo-referencing. The two features that I have found most useful for this project are the Trace Contour and Find Edge filters under the Stylize Filter category. The Trace Contour filter provides a very fine and tunable shadow edge line, but drops out all of the context of the lines. The Find Edge filter generalizes the shadow edge and widens it but it does hold the data in the none-shadow area. The Find Edge form of the aerial photograph greatly aids the fitting of photographs to the shadow bases generated by VNS. Once the Edge Map is generated in PhotoShop, it retains the opaque background of a raster image. To aid in the fitting of the transformed image, it must be converted to a bitmap to gain the transparent background so necessary for fitting to the shadow map. A number

Fig 37: Original OG 82 image Fig 38: Trace Contour of softwares can make the bitmap conversion but it is convenient to do all transformations at the same time.

Fig 39: Find Edge

AutoDesk Raster Design 2004 and Lands Desktop 2004: Raster Design is a very powerful raster image editor for fitting images and Fig 40: Edge Bitmap geo-referencing them in an AutoCad drawing. Raster Design runs in a stand-alone form or as a part of the Lands Desktop which is a special task oriented menu system of AutoCad. When the flight line images have been processed into a bitmap form, they can be easily combined with the shadow base using the transformations in Raster Design. Raster images can also be converted to bitmaps in Raster Design as a part of the insertion process, but several additional steps are involved. Fitting and Geo-referencing the Frame: After the geo-referenced shading base is loaded into AutoCad Lands Desktop, the bitmap form of the frame to be geo-referenced is inserted visually into the scene at the approximate scale and location. At this point the Match function of Raster Design is used to match the Find Edge form of the image to the time matched shadow base. The Match function is a combination of the Scale, Rotate, and Fig 41: Inserting the bitmap Page 15

Move functions of AutoCad. Each match requires two origin and destination points. Raster Design also has a rubber-sheet function, but there are usually not enough distinctive coincident points on a perspective aerial photograph to make this transformation effective. The Match function can be repeated as often as necessary to achieve the level of fit desired. In this case, the bitnap was colored to increase the visibility of the shadow lines for fitting. When the image is fit best as possible with the shadows imaged on Fig 43: VNS image drape the photograph, the corner coordinates of the image can be recorded for use in other applications if necessary. After recording the corner coordi- Fig 42: Fitting the shadows nates, a GeoTiff of the image was prepared in Global Mapper for a perspective drape in VNS for display purposes (Figure 43). As is expected the fit of the image to the DEM has some problems along ridge lines because of the radial displacement in the uncorrected photograph. The same shadow fitting process can be used to quickly create an index of the images along the flight line, another requirement of this project. However, the fit of the images is much easier to achieve in areas of strong shadows. Even though preliminary steps to this shadow fitting process can take some time to accomplish, the actual image fitting process of individual frames can be done very quickly with Raster Design's Match. After the materials were prepared, individual frames could be accurately placed in less than minute, depending upon the amount of time spent refining the fit. As mentioned before, the strength of the shadows have an important influence on the ease of the fitting process. Fig 44: Photo index by shadow matching in Lands Desktop Page 16

Conclusions: Indexing: These experiments, using natural and imaged shadows for indexing, geo-referencing and prediction from Digital Elevation Models of terrain, make heavy use of a variety of software for image processing. The direction for the these experiments were based on the need to index and geo-reference approximately 3000 historical aerial photographs after they were scanned for archival purposes. The photographs are partially scanned and we are now moving to the indexing phase so they can be distributed to the public for use in many applications. Fitting the photographs to period topographic maps was taking from 15 to 30 minutes by conventional visual indexing methods. The use of shadow fitting will greatly speed this process once the preliminary shadow base development is streamlined. A pressing need is the development or location of a pattern recognition software that will automatically match the photo shadows or other terrain detail to the predicted terrain shadows cast from DEMs by Visual Nature Studio or other software. Picture Window, a software used by astronomers and professional photographers for image editing shows potential for matching without having to process the images through as many steps. Automatic fingerprint identification software also shows some potential, but the cost is prohibitive at this time. Shadow prediction: The first set of experiments reported on shadow prediction from DEMs in Nualolo Kai were less conclusive in application potential but show a very interesting direction for future research. With the large volumes of digital elevation data being produced worldwide, data production is far outstripping knowledge of data accuracy. General accuracy metrics are defined and have been applied to elevation data sets, but very little detailed testing has been done either for contour data, derivative DEMs or Shuttle Radar Topographic Mission data, in other than very small samples. I expect that using well corrected orthophotographs, imaged shadows could provide a useful standard for the comparison of the representation of ridged and mountainous terrain as represented in digital elevation data.

Page 17



17 pages

Find more like this

Report File (DMCA)

Our content is added by our users. We aim to remove reported files within 1 working day. Please use this link to notify us:

Report this file as copyright or inappropriate


You might also be interested in