Forest ecosystem management depends on detailed information about forest stand structure. Traditional ground investigation requires vast amount of manpower, resources, costs, and time, hence it is not easy to promptly obtain accurate information by using ground investigation. Airborne laser scanner (LIDAR) is capable of measuring objects at very high density and penetrating gaps among tree crowns, therefore it can rapidly obtain 3D data about forest stands, and can be used for acquiring related data about forest stand structures. This study utilized an airborne laser scanner data set, acquired using Leica ALS40 scanner, of the Yangminshan National Park area to derive digital surface model for the forest canopy. Furthermore, the forest canopy height model was estimated by computing the difference between the digital surface model and a digital elevation model. The canopy height model was compared to the tree heights measured in eleven ground sample plots in order to assess its accuracy. The results indicate that the forest canopy height is generally underestimated by using the LIDAR data. More accurate forest canopy height model can be achieved, if the accuracy of the digital surface model and digital elevation model can be improved.

The airborne laser scanning technique with GPS positioning and INS orientation is useful for height determination of ground points. In principle, no other auxiliary data are needed, and the coordinates of topographic points can be derived directly from the computational results of GPS and INS data. All current developed airborne laser scanning systems apply the method for the height determination. In the paper, we present a block adjustment model for sufficiently utilizing the overlaps between the scanning strips and study the feasibility of the model in order to improve the accuracy of the height determination by airborne laser scanning technique. Our method is applicable for height data which have a continuous first order derivative. The results show that the accuracy at ±20 cm level is achievable.

From the viewpoint of data fusion, we integrate LIDAR point clouds and digital aerial image to perform 3D building modeling in this study. The proposed scheme comprises two major parts: (1) building block detection and (2) building model reconstruction. In the first step, height differences from LIDAR point clouds are analyzed to detect the above ground areas. Color analysis is then performed for the exclusion of tree areas. Remaining regions are restricted by area threshold. After the refinement by shape and texture analysis, building blocks are established. In the second step, we analyze the height information of LIDAR point clouds to estimate the approximate roof-edges and extract the LIDAR points located on the rooftop. Based on the approximate roof-edges, accurate 2D edges are calculated in aerial image and then transformed into candidate roof-planes. Combined with the extracted LIDAR points, the correct roof-planes are selected and projected back to object space as 3D roof-planes. At last, a patented method SMS (Split-Merge-Shape) is employed to generate building models using the 3D roof-planes. LIDAR point clouds and large scale aerial image in Hsin-Chu Science-based Industrial Park were used in the test. A 1/1000 scale topographic map and reconstructed building models measured manually from stereo pairs were used to evaluate the results.

There are two main parts in our research. The first is to segment objects and to produce digital terrain model by using two mathematical morphology methods, i.e. flattening operation and H-Dome transformation, and control posts. The second is to detect buildings by slice the objects’ height, filtering out small areas and computing their statistics of differential slope on region-based method. There are two test areas, one is around Hsin-Chu Science Based Industry Park, and the other is around Ta-Keng area. The results of accuracy for digital terrain model production are -0.16m and 0.04m for mean, 0.45m and 1.12m for RMSE, respectively. And the successful rate for building detection is 87.7%

Lidar point cloud is a collection of a large amount of accurate 3D point data densely distributed on the scanned object surface. It implicitly contains abundant spatial information. To explore valuable spatial information from Lidar data becomes an active research topic. However, the processing of the huge amount of point data needs unusually large CPU time and storage volume. A well organized data structure is needed for the analysis or extraction of spatial features from point cloud in an efficient way. This paper proposes a novel method using an octree structure to organize point cloud in a 3D regular grid. The nodes of octree structure represent the cells of the 3D grids, which can be accessed easily and quickly based on a designed 3D array index. Two 3D plane extraction methods, named split & merge and 3D region growing, are proposed. These two methods fully explore the designed octree structure of point cloud. The plane extraction methods are practicable to both airborne and ground-based Lidar data. Test data include an airborne Lidar data set containing points of multiple echoes and a combined point cloud data set from several scan stations of ground-based Lidar were processed. The test results show meaningful feature extraction from lidar data. This study provides a stepping stone for the extraction of spatial information from lidar data.

DEM generation is the primary application of airborne LIDAR. The point cloud provided by airborne LIDAR not only represents the terrain surface, but also contains buildings, egetation, or other ground objects. The major process of generating DEM from airborne LIDAR is to filter out non-ground points from the point cloud data. The purpose of this study is to propose an adaptive filtering algorithm for DEM generation using airborne LIDAR data. The filtering algorithm is based on the principle of morphological filtering theory. To make the algorithm adaptive, a 3-D grid structure is used to organize point cloud data, so that the trend surface and local slope of the ground can be estimated. The feasibility of the proposed algorithm is tested by using some test data with different characteristics of topography. The algorithm is proved to be effective and practicable in most test cases, but in some cases some ground points in the rough terrain may be over-filtered. To assure the quality of DEM product, manual check and editing is still necessary against improper filtering results.

Airborne and ground-based LIDAR systems are able to quickly support the tasks of three dimensional spatial data acquisition and provide users with abundant geo-spatial information. Optical images, on the other hand, recorded with adequate radiometric detail offer advantages for better interpretation of the scene, three dimensional coordinates of object points can only be determined after orientation and intersection steps, however. Thus, one of the apparent gains in fusing LIDAR data sets with photogrammetry is to solve the photo orientation by using control information resulting from extracting features from LIDAR data sets. This study aims at employing control lines extracted from LIDAR data sets solving for single photo orientation. The functional model describing the geometric relationships among perspective center, line observations in image space, and that in object space, the random model characterizing the control line measuring errors both in image space and object space are established and tested in this work. The experimental results suggest the application potential of fusing LIDAR data sets with images for the task of single photo orientation.

Airborne lidar system integrates the laser ranging, GPS, and inertial navigation systems. This integration provides an instrument capable for high density and high accuracy terrain relief measurement. A pilot study was conducted by the National Chiao-Tung Universtity, with support from the Council of Agriculture. This paper reports the process of this pilot study.The data acquisition includes two commercially available systems, Optech ALTM 2033 and Leica ALS 40. The flight period for the Optech ALTM 2033 is between March 20 and April 3, 2002. Test areas include the Ta-Keng of Taichung, and Jeou-Fen-Ell-San. The flight for Leica ALS 40 is conducted from April 10 to 16, 2002. Test areas include Ta-Keng, Jeou-Fen-Ell-San, Hsin-Chu, Tsao-Ling-Tarn, and Yang-Ming-San. The agreement of lidar measurements and field data gathered with RTK-GPS and total stations differs with land surface cover type. Comparing with the field data in the Jeou-Feng-Ell-San area, the root mean square error of Optech ALTM 2033 is 9.9 cm for the paved road surfaces. There are systematic errors presented in the Leica data set. After removing the average error, the root mean square error is 11.6 cm. In the area of 30% slope and covered with tea plants, the root mean square errors of both systems are less than 38 cm.