This paper presents a new approach to stable automatic extraction of building information from a high resolution satellite image set, panchromatic (PAN) image and multispectral (MS) images shot in a same date. The proposed approach is based on semi-automatic volumetric shadow analysis (SVSA). The proposed approach extracts a height and some information of a same building by SVSA from PAN image and MS images, respectively. The height and some information are used to calculate a maximum cost value by cost functions. A height with maximum cost value is determined as a true height. The true height is used to find a footprint position of the building. This paper also presents an automatic building extraction result by the proposed approach and roof extraction. For test, a Kompsat-2 image set was used. The results showed that the proposed approach can extract building information stably from a high resolution satellite image set.

The current practice in photogrammetry adopts scenebased approach for establishing sensor models of satellite images. One prepares ground control points (GCPs) for a scene of interest, applies bundle adjustments and estimates orientation parameters for the scene. In this paper, we investigate rigorous sensor models for modeling satellite orbits, instead of individual scenes, through bundle adjustment process. Using GCPs from one scene we aim to estimate orientation parameters that can be applied to all images acquired from the same orbit. We selected two types of well-known rigorous sensor models: collinearitybased type and orbit-based type. For each type, we defined seven adjustment parameter sets and tested the performance of orbit modeling for each the sensor model type and adjustment parameter set. We used two strips with a length of 420 km (seven SPOT-3 scenes), which constituted a stereo pair. With each strip, we tested the feasibility of estimating orientation parameters for the whole image strip using GCPs from single scene. We also tested the feasibility of absolute orientation from orbit modeling, i.e. the ability of acquiring 3D information from a stereo strip using GCPs from single scene. Results showed that although there was little difference between the two models for individual scene modeling, for orbit modeling the orbit-based model showed better results. The orbit-based model that used attitude biases as adjustment parameters produced promising performances for modeling the orbital segments and for absolute orientation.

Extraction of man-made structures from satellite images is one of the essential issues in remote sensing and many techniques were proposed for building extraction from high resolution satellite images. However, most of them rely stereo analysis or additional data sources such as LIDAR for retrieval of 3D building information. This paper proposes a semiautomatic approach that extracts 3D building information from single satellite images. In this approach, after measuring boundaries of a building roof we determine the building height by projecting building shadow onto the image space until the projected shadow matches the actual shadow. Then we extract building footprints by translating roof boundaries along the direction of vertical edges to the amount determined by the building height. To increase the level of automation, we devised a technique that adjusts the height so that the projected shadow matches the actual shadow and determine building height automatically. We also devised a technique to extract roof boundaries by clicking only two corner points. A panchromatic IKONOS image and Quickbird image over Deajeon area were used to test proposed algorithms and a IKONO stereo pair was used to get reference building heights. While building heights determined manually showed the accuracy of an RMS error of 1.66m, the heights determined automatically showed an RMS error of 1.86m. The results show that 3D building information can be extracted from single satellite images and that the automation approaches proposed in this paper worked.

The main objective of this paper is to investigate accuracy of various sensor models developed for linear pushbroom-type satellite images. In particular this paper investigates the accuracy of sensor models in two aspects; the accuracy of bundle adjustments and the accuracy of estimating exterior orientation parameters. Analysis of the first accuracy has been reported by many previous studies whereas the analysis of the second accuracy has somewhat been neglected. We argue that the second accuracy is as important as the first one. It describes a model's ability to predict satellite orbit and attitude, which has many direct and indirect applications. We implemented and tested two types of physical sensor models based on the collinearity equations: one with position and euler angles as model parameters and the other with orbit and attitude angles as model parameters. Tests were carried out with two Kompsat-1 EOC scenes and GPS-driven control points. Test results showed that the model with orbit and attitude parameters produced better estimation of exterior orientation parameters while maintained comparable accuracy of bundle adjustments. © 2005 IEEE.

Precision correction is the process of geometrically aligning images to a reference coordinate system using GCPs (Ground Control Points). Many application of remote sensing data, such as change detection, mapping, and environmental monitoring, rely on the accuracy of precision correction. However it is very time consuming, laborious and expensive process. In this research, we propose a new method for automatic precision correction of satellite images using the GCP chips collected from lower resolution satellite images to reduce or eliminate the number of GCPs required. Different Earth observation satellites provide the satellite images of different resolution, different swath and different position accuracy. In general, the satellite images of lower resolution have wider swath and vice versa. In that sense, if we can utilize the GCP chips collected from lower resolution satellite images to do precision correction of the higher resolution satellite images then the number of GCPs required to have a precisely corrected images of the same area would be reduced. In this experiment, we used GCP chips collected from Landsat-7 panchromatic images of 15 m resolutions to perform precision correction of the KOMPSAT-1 EOC images of 6.7 m resolutions. In this case, since Landsat-7 images provide higher positional accuracy of less than 250 m at the systematically corrected level (L1G) comparing about 2 km positional accuracy of KOMPSAT-1, GCPs collected from level 1G of Landsat-7 images can provide enough accuracy in certain applications. To utilize this approach easily, this study would be applied to the automatic precision correction method developed previously, which exploits the normalized cross correlation and the RANSAC (Random Sample Consensus) algorithm to removed the outliers of matching results. The initial result showed the possibility of using GCP chips of lower resolution to carry out precision correction in the frame of the automatic precision correction method utilizing RANSAC algorithm.

We propose a new algorithm for automated registration of satellite images. Registration means finding the relation between image coordinates and a reference coordinate system. The algorithm consists of two steps. The first one is the automated generation of control points. An automated matching based on normalized cross correlation is used. We have improved the accuracy of matching by determining the size and shape of match windows according to incidence and scene orientation angles. The second one is the robust estimation of mapping functions from control points. We used the random sample consensus (RANSAC) algorithm for this step. We argue that it is the second step which gives the robustness of any automated registration algorithms. We carried out experiments with SPOT images over three test sites. Through matching, a number of control points were generated. RANSAC was applied to the control points. All outliers were correctly identified for all three test sites and mapping functions estimated without outliers. The accuracy of estimation was comparable to that of estimation with control points generated all by manual measurements. The results support that our algorithm can be used for robust automated registration.

We introduce a class of path optimization problems, which we call "sweeping path problems," found in a wide range of engineering applications. The question is how to find a family of curve segments on a free-form surface that optimizes a certain objective or a cost while respecting specified constraints. For example, when machining a free-form surface, we must ensure that the surface can be machined or swept as quickly as possible while respecting a given geometric tolerance, and while satisfying the speed and the acceleration limits of the motors. The basic requirement of engineering tasks of this type is to "visit" or "cover" an entire area, whereas conventional optimal control theory is largely about point-to-point control. Standard ordinary differential equation-based Lagrangian description formulations are not suitable for expressing or managing optimization problems of this type. We introduce a framework using an Eulerian description method, which leads to partial differential equations. We show that the basic requirement is expressed naturally in this formulation. After defining the problem, we show the connection between the two perspectives. Using this reasoning, we develop the necessary conditions for the optimality of the problem. Finally, we discuss computational approaches for solving the problem.