Filtering Of LIDAR Data And Extracting Building With Auxiliary Image

The filtering and building detection of airborne lase scanning data is one of the critical data pre-process technologies, so systemic and deep research of the technique has practical value and application background, and it has been a hot topic in current photogrammetry and remote sensing domain. On the base of analyzing the acquiring theory of airborne lase scanning data and characteristics of airborne lase scanning system, the filtering theory for point cloud data is researched and the support vector machine theory is introduced as the classification method of laser points. Then the approach of mathematic morphological filtering based on virtual grid volumetric pixel and detection of building laser footprints based on unbalanced support vector machine are researched, and the experiments have proved these algorithms to be effective and reliable. The mian work of this dissertation are as follows:1. The location theory of LIDAR data and definition and correlation of system parameters are expatiated, which provide theory evidence for practical project application and data post-processing.2. LIDAR data structure method based on virtual grid volumetric pixel is preposed. Different resampling methods for each grid are used according to the characteristics of grid data. So it improves resampling efficiency, avoids to manually creating false data in data blank area, and provides data structure method for back filtering algorithm.3. On the base of virtual grid volumetric pixel for data structure method, an improved multi-scale mathematic morphological filtering method is put forward for adaptive of the method. A small value of terrain grade parameter and fixed filtering parameters are used, and then the quality control method of searching for misclassified ground points can reduce the type I error in filtering results, which resolves the adaptive problem of parameters for multi-scale mathematic morphological filtering method.4.An extraction strategy for building laser footprint is bringed out based on classifying single laser point with support vector machine method. The main idea of the approach is that treate single laser point as classification objects rather than the segementation of laser points, so it can avoid the problems of error transmit of segementation and higher cost of misclassification for segementation of laser points. It also takes the spectrum information and height features of DSM and nDSM as the feature vector. Theory analysis and experiments show that the algorithm has high classification accuracy of building detection, and higher priority than the support vector machine classification method based on only point cloud data. The design of the approach also can satisfy the classification requirements for fusion of different data sources. 5. Detection method for building laser footprints based on unbalanced support vector machine is put forward. Considering that in city area or thick building area, the number of building laser footprints is much larger than tree points, so the standard support vector machine classification method has some problems in such environments. The solution is setting different penalty factors for adjusting unbalanced data. Experiments show the approach to be accurate, credible and precise. Compared with standard support vector machine classification method, it has higher classification precision on both balanced data and unbalanced data, so it can reduce the requirement of support vector machine for data balance and improves the practicability of the approach.