Study On The Technical Issues Of Three Dimensional Change Detection Based On Multi-temporal LiDAR Data

The up-to date character of geospatial information data is the soul of GIS. There is a very urgent need to recognize the importance of real-time detection of the instantaneous variable change in economic construction and national defense construction.With the progress of science and technology and the development of the society, the demand for change detection has also been improved. Meanwhile, real-time detection of the instantaneous variable change using satellite remote sensing data and other auxiliary data is becoming a key technology in land survey, disaster evaluation, environment monitoring and Fundamental geographic database updates. However, in reality there are many problems in the detection of variable change based on Remote sensing images. Detailed problems include heterogeneity within the object, abundant texture information, difference between shadow and projection, decomposition for the mixed pixels and so on. More importantly, the change of geographic data shows up as the change of three-dimensional space, which is hard to be detected by two dimensional remote sensing images in different periods. Light detection and ranging (LiDAR) system, as a newly emerged sensor, has many advantages. This technique supplies a new sounding way with whole-day observation and short-time acquisition of high precision three-dimensional point cloud. The high penetration capacity of laser beam can compensate the block formed by vegetation to some extent, thus reduce data losses of certain areas and the quantity of Feature Collection has been greatly improved. Compared with the detection method based on imaging approach, LiDAR has an unparalleled advantage in the characterization of complex construction and topological relation of surface features. In the other hand, compared with imaging, LiDAR is lack of corresponding semantic Information and spectral information. Because of its high density and large quantity, the point cloud of LiDAR is semi-random discrete distributed. In order to get the three dimensional viable detection with high efficiency and accuracy, there is still some remained key problems to be solved.In this thesis, we focus on the key technology of detection of three dimensional change based on airbone LiDAR point cloud data. The main contents of the research include the following aspects:Study of the airbone LiDAR data pre-processing technology to detection of variable changes. The3D variable change detection based on airbone LiDAR point cloud data is not a standalone processing technique, but needs a series of pre-processing work. In this thesis, we summarized these pre-processing techniques. Two algorithms are focused on, one is the LiDAR point cloud filtering algorithm aimed at complex environment, the other one is the automatic registration algorithm of LiDAR point cloud and remote sensing image. This systematic study of pre-processing techniques will provide significant data for detection of3D changes.Study of3D terrain change detection on the basis of multilayer local ICP matching. Terrain change detection is one of the most important contents in3D change detection, and the key to through the3D terrain change detection is the precise registration of Multi-temporal LiDAR data. The change information between multi-temporal LiDAR data is so rich that traditional global matching algorithm based on least square theory which have the shortage of ability to eliminate noise can not reach a good registration result. A noval3D terrain change detection algorithm based on multilayer local ICP is proposed. The point cloud data is divided into multiple windows through error analysis to determine the optimal window parameters, and determine whether a window contains change information by the difference information entropy. Then the windows with no change information are utilized to match and the optimal model parameters can be obtained and3D displacement on the surface of the earth can be calculated.Study of3D building change detection based on multi-temporal LiDAR data. Buildings are the key elements of urban planning and construction, are also the most dramatic change units in the city. It is very important to achieve the change detection of building structures. In this paper, The advantage of LiDAR data in the complex structure of topological relation on the three-dimensional representation is used to delineate the candidate change regions rapidly, and then the candidate change areas are utilized as constraint, the SVM algorithm are introduced for automatic classification of point cloud. Aiming to the drawback that traditional LiDAR-based classification algorithm of low accuracy, height texture features and local geometric features are extracted to improve the classification accuracy and universality. Finally, the classification results and the spatial analysis results are combined to analyze the building change attributes and high precision spatial characteristics of LiDAR is used to realize building microstructure change detection.Study of3D change detection based on human-computer interaction discriminant mechanism. Due to the technique limitation of many respects, it’s very hard to get full-automatic change detection at present or in the near future. Especially in some particular important area involved with social economy and people’s livelihood, we need to do accurate measurement using manual work on the basis of automatic detection. Airbone LiDAR can get high precision3D topographic data and it needs manual work to analyze the spatial characteristics of topography to determine the changing areas and property. The terrain component render algorithm aimed at low resolution DTM is in current use. It will seriously affected the production efficiency in the use of high resolution topographic data generated by LiDAR data. In order to solve this problem, there is urgent need to develop a good terrain rendering algorithm aimed at the3D change detection of human-computer interaction. In this thesis, we learn from the principle of energy absorption and radiation, study the dSVF rendering algorithms which is not affected by light’s location and terrain fluctuation. In the meantime, we study the strategy combined dSVF characteristics with multi-source feature. By using this strategy, the topographical information of will be more abundant than before. Finally, we discuss the method that using artificial visual discrimination based on auxiliary dSVF rendering results to assist change detection manual work.