Three dimensional building models are the indispensable component in the cyber city. This investigation integrates 2-D maps, LIDAR data, and aerial images for building modeling. This research handles flat, gable, and cambered roofs. Vector maps are used to locate the building boundaries in this research. Since a cambered roof does not have significant features in the image space, we use the LIDAR point clouds to model it. Because the density of the LIDAR point clouds might not be sufficient to reconstruct the internal facets of buildings, we employ aerial images. In the first step, the data preprocessing encloses the polylines of the maps then extract the point clouds that belong to a building. After filtering the point clouds, we fit the data by different surface functions. Through the roof hypothesis by employing point clouds, the camber roofs are parameterized. For non-camber roofs, the ridges of gable roofs will be intercepted by the two inclined planes. The step-edges of flat roofs are obtained by combining point clouds and image features. Then the lines are projected to the object space by ray-tracing. Finally, we shape the models by SMS method. The test site is in the Industrial Technology Research Institute of Hsin-chu. The vector maps are with a scale of 1:1,000. The point density of LIDAR data is 1.5 points/m2, and the spatial resolution of aerial image is 0.12 m. The result indicates the successful rate is 80% in building classification while the fully reconstruction rate is 85%. The RMSE of building boundaries are 0.51 m and 0.41 m in X and Y directions, respectively. The shaping error is 0.19 m.

In this paper, the authors elaborate on an innovative approach for registering terrestrial LiDAR point clouds of overlapping scans based on 3D line features. The proposed working scheme consists of three major kernels: a 3D line feature extractor, a 3D line feature matching mechanism, and a mathematical model for simultaneously registering LiDAR point clouds of multi-scans on a 3D line feature basis. The combination of approximate and rigorous transformation models makes fully automatic transformation possible. Besides, the simultaneous adjustment prevents errors from being accumulative among scans. Experiments conducted show the proposed method of employing 3D line features to be a useful alternative or complement to point, surface and other features for LiDAR point clouds registration. It is especially effective in areas rich in man-made structures.

The distance-dependent GPS errors, notably atmospheric refraction, reduce the success rate of epoch-by-epoch ambiguity resolution, and limit the GPS positioning accuracy, especially for mediumto long-range baselines. Using multiple reference stations to model (or interpolate) the distance-dependent biases between the reference station and a rover can extend the distance or improve the positioning accuracy. The objective of this research is to find out the relative tropospheric zenith delays (RTZD) between different reference stations by using known coordinates and to provide rovers with interpolated corrections for more precise positioning. The project consists of three major steps: (1) finding out the RTZDs between reference stations, (2) modeling the RTZDs; (3) kinematic positioning. In the first step, a pseudo observation equation of RTZD is added in order to reduce the impact of RTZD on ambiguity resolution. In the second step, different model is used according to the numbers of reference stations. These two steps are the emphases of this investigation. The proposed method needs only one to three epochs to resolve ambiguities, so the effects of cycle slip or data gap are not very serious. A multiple reference stations network located in north Taiwan is used in this research. There are six permanent GPS stations in this network. Test result indicates that if the RTZDs between different reference stations can be resolved successfully, and a proper interpolation model is used, it is possible to improve the positioning accuracy.

Digital Elevation Models (DEMs) are widely used in many spatial information related applications. There are a variety of technologies developed for DEM generation, such as digital photogrammetry, airborne LiDAR and InSAR. However, the procedure from raw data sources to DEM products may involve different types of errors, which determine the DEM quality. Therefore, methods for DEM quality assessment are needed for users. This study proposed two methods to describe DEM quality—interior and exterior relative quality assessment. Instead of whole index of RMSE, the former provides accuracy index of each height point calculated from the raw data to reveal the DEM quality in a local area. Then the later explores the relative errors between DEMs acquired from different sources and times. The two methods provide useful information to DEM quality assessment, which makes a user to realize the quality of acquired DEMs in detail.

Because of the high dimensionality of hyperspectral images, traditional statistics-based classifiers cannot be directly used for the hyperspectral image classification with limited training samples. The commonly used method to solve this problem is dimensionality reduction. Several research works have proven that wavelet transform provides an appropriate and effective tool for spectral feature extraction. Also, the researches about back-propagation neural networks for image classification have been investigated. In this study, wavelet networks is proposed for hyperspectral image classification. With wavelet networks, the important task of feature extraction is performed with wavelet transform, whereas classification is carried out by back-propagation neural networks. Both the position and the dilation parameters of the wavelets are optimized as well as the weights of the network during the training phase. The experiment results suggested that wavelet networks can be a useful and effective classifier for hyperspectral image classification.

Digital aerial cameras can take highly overlapping aerial images with the end lap percentage of about 85% ~ 90%. These images have less occlusion areas. For aerial photogrammetry, such aerial images possess a good application potential for automatic extraction, tracking and matching of feature points. This paper utilizes the successful tracking ability of the iterated optical flow approach within the tracking distance of less than 30 pixels, and develops a new method for image matching and determining the corresponding 3D points. Test results verify that it can determine about one point per 1.5m2 (≈9×9pixels). Also, it can automatically determine one point per 6.8m2 (≈20×20pixels) and one point per 0.2m2 (≈3×3pixels) in occlusion and tree areas, respectively. Moreover, those matched points are checked by aero triangulation. For example, totally 7473 tie points on 6 images are regarded to be successfully matched from 6 overlapping images in a strip by the proposed method. Only 50 tie points are mismatched points detected by reliability theory. In other word, there is about 99.3% of the matched points which are correct and determined by the method. This study proposes the new method for image matching using highly overlapping aerial images. It provides more redundancy and more reliable matching results than the ones done by traditional stereo pair. A huge number of matched points can be then applied to determine a dense cloud of 3D object points for later applications such as DTM determination and 3D cyber city modeling.