An approach to line photogrammetry using four independent parameters for the representation of a 3D straight line is discussed. The developed algorithm is designed for the reconstruction of objects in map updating from large scale aerial imagery. It allows the introduction of 3D geometric constraints on single and multiple lines in a simultaneous adjustment process. The approach is extended to digital line photogrammetry, by automatically extracting straight line segments with image segmentation methods. The identification of corresponding lines and the 3D constraints are currently provided by the user. In an empirical test it has been shown that line photogrammetry is at least as good as conventional point photogrammetry concerning accuracy and reliability, but with digital line photogrammetry offering a potential automation of the measurement process. Current research involves the development of a multi stage comparator for interactive mapping with a core of line photogrammetry and an automated extraction of straight lines. The final intention is however, to provide the line correspondences and the constraints by automated parsing of image segmentations and to use digital line photogrammetry as a final, precise adjustment process to derive coordinates of lines and corners of buildings with objective quality estimates.

Point- and line-based invariance methods are used for both image and object transfer 2D planar objects. Invariance yields equations of straight lines the intersections of which give the positions of the points to be transferred. For non-redundant 4-point invariance, the sequence of points used yields line pairs of different geometric strengths. In redundant cases, using different point sequences to form linear condition equations results in least squares estimates which are different for both the positions and their quality. Corresponding photogrammetric techniques based on point and line features, on the other hand, provide unique estimates and covariances for both non-redundant and redundant cases. A refined least squares approach, for which the linear invariance equations become non-linear, appears to alleviate the non- uniqueness problem. Point-based image invariance is investigated for 3D objects in multiple images. The concepts of the essential and fundamental matrices for use with calibrated and uncalibrated cameras, respectively, are presented. The use of the fundamental matrix to transfer images from two photographs to a third is described and early results summarized. Introducing a constraint on the fundamental matrix stabilizes the solution, which otherwise leads to widely varying results. Preliminary results from linear invariance based object point transfer suggest that the uniqueness problem arising from point sequences also exists in this task. Research is continuing to resolve this issue and to provide photogrammetric equivalent and complimentary methods to invariance.

The inclusion of geometric constraints in a bundle adjustment greatly improves the process of generating accurate, detailed site models, increasing both the precision and the reliability of the solution. This is especially important for our application, using multiple oblique images to automatically generate detailed site models. We have implemented a bundle adjustment which can include constraints between any of the parameters. The program generates precision and reliability statistics for all of the observations, which, since it is based on the unified least squares model, also include the parameters and constraints. This allows evaluation of the added geometric information for blunders. This paper discusses the mathematics and implementation of the program and shows examples of editing both point measurements and constraint information.

In this paper, we address some issues that arise in calibrating cameras and estimating object points in connection with building a detailed geometric model from multiple images. These issues concern the numerical algorithms for estimating the parameters, the statistical evaluation and validation of the results, and the benefit of adding passpoints in the calibration.

One of the major driving forces behind digital photogrammetric systems is the continued drop in the cost of digital storage systems. However, terrestrial remote sensing systems continue to generate enormous volumes of data due to smaller pixels, larger coverage, and increased multispectral and multitemporal possibilities. Sophisticated compression algorithms have been developed but reduced visual quality of their output, which impedes object identification, and resultant geometric deformation have been limiting factors in employing compression. Compression and decompression time is also an issue but of less importance due to off-line possibilities. Two typical image blocks have been selected, one sub-block from a SPOT image and the other is an image of industrial targets taken with an off-the-shelf CCD. Three common compression algorithms have been chosen: JPEG, Wavelet, and Fractal. The images are run through the compression/decompression cycle, with parameter chosen to cover the whole range of available compression ratios. Points are identified on these images and their locations are compared against those in the originals. These results are presented to assist choice of compression facilities after considerations on metric quality against storage availability. Fractals offer the best visual quality but JPEG, closely followed by wavelets, imposes less geometric defects. JPEG seems to offer the best all-around performance when you consider geometric and visual quality, and compression/decompression speed.

Nonlinear modeling and solution techniques of array algebra are applied to the problem of simultaneous graph matching and photogrammetric bundle adjustment. Graph matching provides automatically the image coordinates and 2 X 2 weight matrices of `control entities', points and vertices of known relative geometrical shape which replace the control points used in traditional bundle adjustment. Inclusion of multiple control entities to cover the entire image area of interest allows the use of a fast new array algebra formation of real-time bundle adjustment to act as the pull-in mechanism for the global graph matching process. The resulting integrated solution of Feature Entity Least Squares Matching (FELSM) is very fast and produces high quality results. FELSM has demonstrated solutions to several problems of ongoing research interest in photogrammetry and the related fields of image understanding, pattern recognition and computer vision. These results open the way for further integration of the various fields.

We discuss Texas Instrument's approach to the Model Supported Positioning (MSP) Project. There are two main goals of the MSP project. The first is to meet the model-to-image registration needs of the Research and Development for Image Understanding Systems (RADIUS) Project. The second goal is to provide an automated capability for deriving more accurate absolute image position information using a scene model with accurate ground control positions. The approach is to use model-to-image registration to locate features in the image whose absolute 3 - 4 positions are precisely known from the model. Rigorous photogrammetry is used to refine the image acquisition parameters by minimizing the error between the known 3D positions and the image derived positions. While we briefly discuss the parameter refinement step, the emphasis of this paper is on the registration algorithm. We first estimate a coarse offset error vector between the image and the model. This estimate is from a vector registration algorithm which matches straight lines extracted from a reduced resolution image to lines from the site model. The image location of each 3D control feature is then determined using oriented line and vertex operators in the full-resolution image. A least-squares approach is used to refine the image acquisition parameters to match the known control feature positions.

There is considerable activity at present in feature extraction using a knowledge based approach. Much of this work is concerned with identifying the features. In the work described in this paper the main interest is in extracting features which can be matched on an image and a map. The emphasis is on polygons which have a distinctive shape and can be extracted from vector as well as raster data. An algorithm has been developed and tested which automatically matches forest areas on a map with forest areas on and image in two dimensions. This algorithm is being developed for more general use so that the problem of image to map registration can be automated in a general way. Results are given on the use of the algorithm to identify changes in forestry from Landsat Thematic Mapper data and developments which can match buildings on high resolution satellite imagery are reported. An overall strategy for full automation is presented and discussed and future developments considered.

A framework is presented for the interpretation of a urban landscape based on the analysis of aerial pictures. This method has been designed for the use of a priori knowledge provided by a geographic map in order to improve the image analysis stage. A coherent final interpretation of the studied area is proposed. It relies on a graph based data structure to modelize the urban landscape, and on a global uncertainty management to evaluate the final confidence we can have in the results presented. This structure and uncertainty management tend to reflect the hierarchy of the available data and the interpretation levels.

Observation of interferometric images provided by Synthetic Aperture Radar satellites enables the reconstruction of Digital Elevation Model of an area. The elevation of a ground point is computed from the interferogram which is the phase difference of two signals backscattered from the same point and emitted from two different positions of the satellite. Since the phase difference may only be measured module 2 (pi) , the problem consists of estimating the unwrapped phase for each point, that is the number of complete cycles of the phase. Here, we present a solution for this problem which is based on a Markovian approach allowing a combination of global and local constraints. As in regularization technics an energy function is defined as the sum of two potential functions. The first potential measures an error with respect to an a priori model (here, the membrane model). The second potential measures an error with respect to the data (i.e. the interferogram). The optimal labeling with respect to this non-convex energy is then computed by a simulated annealing process. The computational cost of this relaxation technique is particularly heavy in phase unwrapping. This is due to the residual gray level difference between the unwrapped phase image and the data. Therefore, we first employ a deterministic algorithm, which is faster but may propagate some errors, before applying our stochastic algorithm which should correct most of those errors. This algorithm was implemented on the Connection Machine CM-200 and applied on numerous examples presenting cases of discontinuities. The results are quite satisfactory from many points of view: robustness, sensitivity to noise and presence of discontinuities.

When a terrain elevation map is computed from widely separated images the perspective distortion may result in a large number of false matches and poor reconstruction accuracy. This paper describes three image matching algorithms designed specifically to process images taken with large base-to-height ratios. They include a new match score, a subpixel interpolation scheme, and a multi-resolution unwarping technique. The algorithms are incorporated into a stereo analysis package and the system is tested by processing a single pair of high altitude images with a base-to-height ratio of 0.63 and a sequence of simulated images with base-to-height ratios that varied between 0.25 and 2.25. Analysis of the simulated data show that when these techniques are implemented the reconstruction accuracy remains independent of the base-to-height ratio.

The SAR interferometric technique, in particular using the ERS-1 data, already proved to be an effective method to extract DEM. However, the detail analysis of the accuracy and artifact level in the generated DEM tends to lower the initial expectations and reveal the need, far more than for optical stereoscopy, to understand the imaged scene. Actually some scene analysis is necessary to desambiguate phenomena conveyed by the phase signal although in return these informations may improve our understanding of the scene. Therefore we will present the use of a priori knowledge in the SAR interferometric DEM generation and how the interferometric product can then be used to improve our scene analysis. In particular we show here that from interferometric products one can generate good quality geocoded radar orthoimages which then allow easier scene interpretation.

The U.S. Geological Survey is converting its primary map production and revision operations to use digital photogrammetric techniques. The primary source of data for these operations is the digital orthophoto quadrangle derived from National Aerial Photography Program images. These digital orthophotos are used on workstations that permit comparison of existing vector and raster data with the orthophoto and interactive collection and revision of the vector data.

This paper presents the research activity of the French National Geographic Institute (IGN) in the field of digital photogrammetry. Focusing on aerial stereo pairs analysis, IGN research policy was set up in 1990 in the specific context of the production of a topographic database on the whole national territory. Stress was given to four major applications: contour line generation from automatic correlation techniques, road semi-automatic plotting, 3D building extraction, and map analysis. This latter topic actually started in 1994. The other three ones were undertaken in 1991 within a four people team, that was to grow up to 15 in 1995. Nowadays, IGN come to a key point of this activity, first because the primary results of the research are reaching enough maturity to foresee possible operational use, and secondly because the production lines are now evolving toward the digital stereoplotter technology.

The Central Imagery Office, established in May 1992 as a Combat Support Agency of the Department of Defense, has begun implementing a range of initiatives designed to enhance the capabilities of the United States Imagery System and improve the overall U.S. imagery effort in response to continuing shortfalls amplified principally during Operations DESERT SHIELD and DESERT STORM in the Persian Gulf. One of these initiatives, and the subject of this paper, is the Exploitation Process Reengineering Study, or ExPReS. Focused on the area of imagery exploitation, this initiative aims to ensure all aspects of this particular process are sufficiently robust and effective to stay in step with other, rapidly-expanding segments of the imagery cycle.

A system for automatic ground control point measurement and rectification of satellite images to a nationwide reference database has been developed. As reference for the rectification a database consisting of more than three hundred thousand features covering the whole Finland has been created. These features are islands and lakes extracted from the nationwide Land Cover Classification, produced from Landsat Thematic Mapper (TM) images. Also the use of digital maps and image mosaics has been tested. Lakes and islands are extracted from the satellite image to be rectified. Their attributes are compared to those in the reference database. Using feature based matching and robust estimation a few hundred ground control points of subpixel accuracy are selected to the rectification. Images of different resolution can be measured automatically using this system. It has been tested using SPOT, Landsat TM and NOAA AVHRR imagery. The search for ground control points takes only a few minutes per satellite image. The accuracy of the result has proved to be at least as good as when measuring the control points manually. The method is tested by computing the parallaxes between the reference features and the rectified images.

A digital photogrammetric workstation system is developed in P.R. China which performed automatic mapping from space imagery data efficiently. Since the earlier periods of space programs, surveyors and cartographers have been looking forward to the day when imagery data obtained from space can be utilized in mapping. With the launched and launching satellites, such as SPOT, MOMS, ERS-1, Radarset, etc., especially the announcement of Eyeglass, people are more interested in mapping using space imagery directly. This paper first compares two stereo imagery mapping methods (traditional photogrammetry and digital photogrammetry), then discuss the theory and practice in details in stereo viewing and automatic stereo measurement (image matching) of stereo imagery by Digital Photogrammetric Workstation.

Emerging requirements for diverse spatial data bases to support a broad spectrum of modeling and simulation activities can profit from advances in the state-of-the-art for automated cartographic data base generation and maintenance. This paper provides an introduction to Advanced Distributed Simulation (ADS) and the associated requirements for Synthetic Environments to support Synthetic Theaters of War. It provides a technical rationale for exploring applications of image understanding technology to automated cartography in support of ADS and related programs benefiting from automated analysis of mapping, Earth resources and reconnaissance imagery.

Generation and representation of synthetic environments is a key factor in using a simulation system for the particular application it is designed for. Realtime networked simulation places additional constraints on the successful creation of these environments. When these networked simulators operate in close proximity to the ground, the constraints on the synthetic database become even more stringent. This paper describes some of the challenges associated with the creation and usage of these synthetic databases for ground and near-ground applications in a networked simulation system. The representation of the information for the same terrain region can be drastically different based on the needs of a simulation platform, or the specific application. The paper touches on these diverse computing needs and on how they affect the creation and exchange of a synthetic terrain database in a networked heterogeneous system. For example, creation of electronic or paper maps demands a data representation scheme that may not be suitable for a thermal sight simulator, even though they are both intended to depict and operate in the same geographical region. The key steps in creating synthetic databases are reviewed and the trade-offs in source data selection, geometric representations, and issues such as interchange and interoperability are highlighted. Some management considerations in improving the process, quality, and tool development are also discussed.

An algorithm that significantly reduces the time required to photogrammetrically extract valid double-line drains is described. The algorithm automatically adjusts an initial double-line drain delineation for accuracy and consistency. It is based on a rigorous, optimal solution technique and is currently implemented on the Defense Mapping Agency Digital Production System.

In this paper we describe a new multi resolution approach to automatic road extraction from aerial images. We make use of the fact that different characteristics of objects such as roads can be best detected in different scales. Two different resolutions of the same image are used, a coarse one with 2 m per pixel, and a fine one with 0.25 m per pixel. In the coarse resolution roads are modeled as bright lines and are extracted by a combination of local and global thresholding. In the fine resolution roads are assumed to have two parallel edges, be bright, and have a homogeneous texture. A multi step procedure has been designed to find the roads according to these criteria. Subsequently both outputs are merged using a rule based system. The developed method has been tested on real imagery, and some preliminary results are reported. Based on the existing experience the multi resolution approach is claimed to be superior to a road extraction in one resolution only.

Scanned maps contribution to the aerial images interpretation allows to consider as possible the complete resolution of the geographical objects depicted in a scene. But as all symbolic documents, maps are submitted to unavoidable distortions during their conception process. We present a method to estimate and characterize these alterations in the road network case. An a priori model of the distortions is first built, based on the cartographic drawing rules. Then, a control of the map data is operated with a ground reference given by an aerial image. Finally, a prediction model is established from the results. It gives the roads aspects in the image with tolerance thresholds and alerts the operator when the map has a low reliability.

A system has been developed to acquire, extend and refine 3D geometric site models from aerial imagery. The system hypothesizes potential building roofs in an image, automatically locates supporting geometric evidence in other images, and determines the precise shape and position of the new buildings via multi-image triangulation. Projectively warped image intensity maps are associated with the faces of each recovered building, allowing realistic rendering of the scene from new viewpoints.

Image analysis as an help for the production of cartographic data must obey some constraints in exhaustively and accuracy that are imposed by the specifications of the production. In the methods presented by researchers, exhaustivity is often tested but accuracy, rarely. For the purpose of building extraction, planimetric and altimetric accuracy are required, we build an hybrid method based on two studies that provide planimetric accuracy--a monocular analysis based on perceptual grouping--and altimetric accuracy--a local statistical analysis of the results of a stereo-matching process. We assessed the accuracy of this method on a class of rectangular buildings and compare our results to the accuracy of the first method that inspired us.

The built environment constitutes a very important target for automated image understanding systems. Currently, these environments only have manually surveyed 2D vector information in paper or digital map form. Increasingly, there are many requirements for 3D information. Automated systems are the only viable way to acquire this information over vast areas with a temporal frequency which will keep pace with the rate of change of many of these areas. Most published image understanding techniques for the automatic extraction of man-made objects only address a specific class of scene at a specific resolution. To address the full range of circumstances which will be required and make a technique more robust, it should be applied to multi-resolution images. In this paper, two automated systems one for building height extraction (stereoscopic) and one for building detection from monoscopic images are applied to multi-resolution aerial and spaceborne imagery. These systems were originally developed with for 0.15 m resolution inner city urban area imagery. With 0.24 m resolution suburban imagery, they performed very successfully. With 0.85 m resolution urban imagery containing very complicated buildings, they show promising results. A 2 m resolution Russian DD5 image was also tested with the monoscopic building detection system and the results showed automatic extraction of large industrial buildings is possible with such imagery. In summary, it was shown that these fully automated systems can handle images with various resolutions and environments.

In previous work we presented a system, MULTIVIEW, used to generate 3D building hypotheses starting from sparse features, hypothesized building corners, extracted from multiple views of the scene. This technique results on knowledge about the imaging geometry and acquisition parameters to provide rigorous geometric constraints for the matching process. The effectiveness of this approach was demonstrated using complex aerial imagery taken with highly oblique views containing buildings with flat or peaked roofs. The MULTIVIEW system uses multiple views based on the successive incorporation of new image data into an existing partial solution. This approach allows the generation of data not seen in the initial images and increases the 3D positioning accuracy of derived object models by simultaneous solution of the collinearity equations. In this paper we provide a detailed performance analysis on the 3D buildings constructed by MULTIVIEW. We evaluate these results with respect to several issues including: (1) Metric accuracy of building recovery. (2) The ability to improve detection and delineation as the number of views are increased. (3) The effect of image processing order on building detection and delineation. The implications of incremental construction of detailed 3D structures are examined with respect to manually derived ground truth data.

Terrain analysis in support of planned military training or operations in a task which requires considerably training, skill, and experience. Military planners must synthesize knowledge of both their own and their expected adversary's tactics, weapons systems, and probable courses of action to determine key terrain, those portions of the terrain surface which have the most impact on the conduct of tactical operations. Many attributes of the actual terrain influence terrain analyses. These include elevation, intervisibility, vegetation cover, transportation networks, waterways, trafficability, soil types, and others. In some important areas of the world, the large set of attributes that influence terrain analysis is greatly reduced. Desert areas comprise one such areal class. As an example, a high resolution digital elevation model is sufficient to support most terrain analysis efforts for platoon and company operations in the US Marine Corps' dismounted infantry training area at 29 Palms, California. The digital elevation model allows an analysis to characterize each point in the model according to an approximate relative-visibility metric. Determination of key terrain, siting of probable defensive positions, and identification of highly concealed avenues of approach flow from examination of the resulting visibility model. These tactically significant areas can be used to conduct operations planning, perform DEM resolution studies, or help determine selective fidelity parameters for TIN modeling purposes.