Terrestrial laser scanner can rapidly acquire accurate and dense 3D point clouds covering the surfaces of scanned objects such as buildings. The point clouds provide the detailed data necessary for accurate building modeling. In order to acquire complete data points on a scanned building, the scanning operations must be done at more stations. Each station has its own coordinate system representing the 3D position of each laser point. Therefore, all coordinate systems of different laser scanning stations must be transformed into a common system to register laser points acquired on different LIDAR stations. This paper proposes a semi-automatic method for registration of terrestrial laser point sets acquired on different stations. Firstly, a point cloud on a local plane is selected manually, and then a mathematical plane is fitted in a least squares manner onto them. Three local planes on an object corner are thus respectively determined, and their intersection point is computed by solving these three plane equations. Such points are used as tie points for transforming different laser coordinate systems into a common one. The transformation is bad in cases of inaccurate, or insufficient, or worse-distributed tie points. This paper suggests to apply suitable “virtual corner points” to solve this problem. The test results show that suitable virtual corner points really can strengthen the geometrical configuration for the transformation and thus raise the accuracy of corresponding transformation parameters with the improvement rate of 36% to 71%. On the other hand, the ratio of the RMSD-value to the scanning distance S≥50m becomes a constant of 0.00015 in case of ,Swhere RMSD denotes the root mean square value of the perpendicular distance from a laser point to its corresponding plane.

Cyber city is getting important. Building models, among others, could be the most important elements in a cyber city. Due to its maturity, LIDAR data has demonstrated profound potentials in fully automatic building reconstruction. LIDAR data contains plenty of height information, while vector maps preserve accurate building boundaries. From the viewpoint of data fusion, we strive to integrate two data to perform building modeling. The proposed scheme comprises three major steps: (1)preprocessing of LIDAR data and vector maps, (2)segmentation and detection of wall and roof faces, and (3)building modeling. In the preprocessing stage, the height variation of the above-ground objects is determined by subtracting the surface elevation from the terrain. The closed polygons for buildings are also obtained. In next stage, segmentation and detection of wall and roof faces is implemented by region growing. In the step of the building modeling, the construct edges of a building could be obtained. Then the 2D topology of building edges is obtained by SMS method. Finally, building models are reconstructed. The test data covers Tai-Chung city in Taiwan. The average density of LIDAR data is about 1.71 points per square meter. The vector maps are with a scale of 1:1,000. About 90% buildings are correctly reconstructed by the proposed method. The shaping error is about 0.17m.

Airborne three-line-scanner images have the merits of high spatial and spectral resolutions and excellent converging geometry. Thus, the images have become an important data source in environmental remote sensing and three dimensional positioning. The three-line-scanner acquires three different direction images, which have plenty of overlapped area in flying. Thus, the bundle adjustment is selected to calculate the dynamic orientation parameters. To compensate for the systematic errors, additional parameters are included in the adjustment. The orientation parameters are modeled using low-order polynomials with respect to time by employing GCPs and tie points. The process might contain local systematic errors for the data with high dynamics, thus, a least squares filtering that performs orbit collocation is included. The test data include six strips. In the bundle adjustment, the tie points are selected for each of the three-look images that appear in the overlapped area of strips. The experimental results indicate that the bundle adjustment could reach higher accuracy when the cross strips are employed.

In the research of land-use information survey and monitoring broad and simultaneously, traditional land surveying methods, such as field surveys, are time-consuming and costly. In the past, per-pixel classification algorithms were used for research, which did not consider about the location of pixel within the image and the relationship between its spectral response and neighborhoods. So, it can’t classify land use categories accurately. So, in this research on classification of land use categories from high resolution satellite images, we will use a segmentation method to improve image classification accuracy and make up the shortage on the other methods. In this method, on the one hand we considered three factors of gray level, pixel size, and the relationship between neighbor images, and on the other hand we use two automatic threshold selections for segmentation. Therefore, in this research, we utilized image segmentation to improve the classification results of land use categories from QuickBird satellite images. Then, we composed the program to extract the information in the image space for the application of practice. Finally, we obtained good image classification accuracy (overall = 84 percent; Kappa coefficient =0.81) by using degree of fitting threshold selection for segmentation in QuickBird satellite images and segmented image automatically.

Cartographic generalization represents the process of simplifying a graphic object by reducing the number of its data points. The graphic reduction technique is capable of removing points on a smooth-curve segment, while retaining sharp turning points. Therefore, a good generalization algorithm is not only capable of simplifying data points, but it also retains the similarity of the generalized curve to the original one as close as possible. Among many generalization procedures, polygon shape generalization is considered the most complicated. Since most polygon generalization methods simplify the polygon as poly-lines, therefore the generalized result is significantly affected by the decision of the starting point, which is required in initiating the poly-line generalization algorithm. The purpose of this study is to develop a curvature-based generalization approach for polygon GIS data. Because the curvature can be used to quantitatively measure the shape of a curve, therefore, this approach uses curvature as a generalization index. According to the curvature property, this study classifies the data point of polygon into two types: the critical points and the secondary points. The critical points are the points with relatively large curvature (often with sharp angles) and can be detected automatically by using a curvature threshold. Then the critical points will be used to divide the closed polygon into the curve segments, which can be employed to extract the secondary points. The extraction of the secondary points is performed by ranking the significance of the data points on the curve segments. The significance of the data point will be decided by the curvature and its relative positions to other data points. The data points with larger curvatures and longer distances to neighbor points will have higher chance to be selected as the secondary points, because after simplification, they will not change the shape of the polygon largely. An experiment is performed to compare the proposed method with the Douglas-Peucker method. The test shows that the generalization results of the proposed method have fewer shape distortions, and it can be used to provide an efficient and automatic tool for polygon GIS data generalization.

The study uses Rational Function Model(RFM) to perform the correction for the Formosat-2 image,because it must use a minute constant h to stabilize the computation of normal equation at high degree RFM,this research would investigate the effect of h value;finally,it would also check and analyze the residual. The test area of Formosat-2 image is Tainan, image level is Level 1A,image size is 12000pixel×16800pixel.The research would use different areas and different number of ground control points with different distribution to test, and change the constant h when every test to get the best constant h. The experimental analysis demonstrated,when uses at least 7 peripheral distributed GCPs to compute, the RMSEe of check point is ±1.50(pixel) and RMSEn is ±1.75(pixel).The proposed h value for second degree RFM and third degree RFM are 0.0002 and 0.00045 .Eventually，the result of t-test doesn’t demonstrate systematic residuals ，when α=5%.

Watershed segmentation is a widely used technique for image partitioning. However, over-segmentation normally occurs when the smoothing process is not operated properly. For example, improper selection of smoothing scheme and parameters will not only wipe out the detail of image content but also affect the accuracy of following feature extraction. In order to overcome these drawbacks, a modified watershed segmentation method based on the edge intensity filtering is proposed. An experiment using both QuickBird multi-spectral and panchromatic images is performed to test the proposed scheme. The result shows that the proposed scheme can generate reasonable and simplified segmentation output for different type of objects.

This study focused on the DEM comparison for two InSAR DEM data. One is the Shuttle Radar Topographic Mission (SRTM) 3 arc data, and the other is Topographic Synthetic Aperture Radar (TOPSAR) data. The DEM produced with airborne photogrammetry is used to validate the system bias, random error and blunders of those InSAR DEMs. The accuracy of SRTM data is 9.5m in the mountainous area and 7.6m in the flat area, based on the evaluation conducted. It is better than the official specification. But the TOPSAR data that contained massive blunders was not as good as the system specification for either the flat or mountainous area. The results show that the unedited TOPSAR data is hard to achieve the nominal accuracy of 1~5m. After removing the blunders, the standard deviation of DEM difference is about 5.6m in the flat area and 12.4m in the mountainous area.