Research On Objects Classification And Simple Buildings Reconstrcution Based On Digital Images And LiDAR

LiDAR, Light Detection and Ranging, provides a new technological means for attaining high-resolution and high temporal-spatial geo-information, which makes data acquisition are being changed from the traditional photogrammetric mode to the continuous automatic mode. It enhances the accuracy and speed of survey, and automate acquisition and processing of the data. Since 1990s, LiDAR technology has been widely used in the fields of object extraction and surface feature extraction and 3D reconstruction. There are still many difficulties in object extraction and building reconstruction only from point cloud, because data provided by LiDAR system is discrete and regular distribution, and lacks spectrum, texture and shape information. in addition, the data gap also exists in LiDAR data. However, with the improvement of hardware techniques, LiDAR system generally includes high-resolution digital camera, from which the high-resolution aerial images, even infrared images, can be attained when accessing the point cloud. The aerial image provides aboundant spectrum and texture information that can compensate the disadvantages of the data of LiDAR. It is a tendency to the field of Photogrametry and Remote Sensing that the shortcomings of one data source can be made up other data sources through the integration of different data sources. This paper studies a few of problems and key techniques in the object classification from integrating LiDAR and aerial images, as well as in the process of building extraction from the following aspects:The several techniques of quick inquiring the discrete point cloud have been introduced respectively, which include the two-dimensional grid index, the quick query of K-nearest and region neighbors in the three-dimensional by using KD-tree structure, and Triangular Irregular Network(TIN). Moreover, several key techniques in using the filtering algorithm based on Progressive TIN densification, which consists of selection of seed points, division of grid cell size removal of outliers, and selection of height threshold.The two-stage filtering method based on scan line has been presented according to the characteristic of LiDAR data aligning scan line. At the first stage, based on the feature of continuous terrain, terrain points can be extracted in one-dimension by using the slope height difference of two neighbors and the maximum terrain slope of scan region. At the second stage, the terrain points extracted from the first stage refer as the candidate points, from which the non-terrain points will be excluded utilizing the local parameter surface fitting in the assumption that urban local terrain surface is flat. Experiments show that the method is the feasibility of urban terrain.The object extraction of fusing LiDAR data and aerial images has been proposed by analysis of the spectral information and geometric information provided by the aerial image and LiDAR data respectively. Firstly, the homogeneous regions of image can be attained by the pyramid spliting-merging segment algorithm and boundary tracking which can be treated as the object of classification. Then, line features extracted from the aerial images, geometric information, such as height information, discrete measurement, and the height difference between the first echo and the last echo, provided by the LiDAR data, which all refers as the clues for classification. At the last, in the each homogeneous region, the spectrum information and geometric information can be comprehensively used for the classification of LiDAR data in accordance with the experiments to determine the conditions and rules of classification.An algorithm for simplification of building region polygons, the Least Square Template matching and constraint of right angle for extraction of simple righ angle building and the automatic segmentation for the roof of building has been prensented. On the basis of using RANSAC algorithm to detect the surfaces of roof, the ridge points of gable building can be reconstructed through building the neighbor correlation of surfaces of the roof, from which the ridge lines of roof can be attain. Due to disadvantages of LiDAR data, the boundaries of building are not real meaning of them. Finally, the building models reconstructed from LiDAR data need further to locate accurately by using the advantages of aerial image.In this thesis, many filtering experiments had been conducted for testifing the two filtering algorithms, and their results show that the proposed algorithms are completely feasible. In addition, a few of key techniques had been discussed. Because this study is on the basis of already registrated between LiDAR data and image data, research on registration does not introduced. Through lots of classification experiments, problems and possibilities in practice are discussed. Due to limitation of resolution of images, only three simple building types had been reconstruction. Object classification and model reconstruction, based on integrating LiDAR and Image, are a complicated task. How to improve the accuracy of classification and automation of building reconstruction still need further study.