Study on Data Filtering and Building Extraction of Airborne LIDAR Data
|School||PLA Information Engineering University|
|Course||Photogrammetry and Remote Sensing|
|Keywords||airborne LIDAR filtering building extraction slope filtering GLCM texture features NDVI clustering algorithm|
Airborne LIDAR (Light Detection and Ranging) orientates terrain’s positions directly and acquires terrain surface’s geo-coordinates and characteristics by scanning the terrain with laser. As a newly emerged instrument of terrain mapping, not only airborne LIDAR solves the problem of mapping difficult regions by traditional aerial photogrammetry, but also puts some influence on the development of terrain mapping. The aim of this dissertation focuses on two tough problems: point clouds filtering when obtaining digital elevation models (DEMs) and building extraction when modeling urban buildings with airborne LIDAR data. A filtering algorithm called slope slef-adaptive filtering and an automatic building extraction strategy based on airborne LIDAR data are presented in this dissertation.The primary works and innovations are included as:1. The components of airborne LIDAR system, LIDAR data’s characteristics and LIDAR system errors are introduced firstly, and then the airborne photogrammetry technique is compared with airborne LIDAR. Some classical point clouds filtering algorithms are reviewed, and by analysis and conclusion, these algorithms are sorted into three types: surface based, region based and slope based. The methods and algorithms used in building extraction are also reviewed, and a few problems that need to be settled are summarized.2. A slope self-adaptive filtering algorithm is put forward. The idea of this algorithm is from CAS filtering model, but makes some improvement. There are four types of slope thresholds in this algorithm, which are general slope, slope increment, minimum slope and maximum slope. All these slopes are used as constraint condition when searching for ground points, thus, overcome the disadvantage of searching for ground points only with slope and slope change thresholds. The terrain’s slope is estimated priorly, so the filtering algorithm can acquire more accurate slope thresholds. The experiment result validates the algorithm being self-adaptive and stable.3. An automatic building extraction strategy is presented that uses first echo of airborne LIDAR point clouds and multi-spectral image. Characteristics of buildings in the LIDAR point clouds are learned first. Some knowledge or theory, such as mathematical morphology, digital image processing and pattern recognition are also utilized in processing. For each building candidate, the gray level coorcurrence matrix (GLCM) texture features are extracted, buildings and trees are identified by clustering algorithms using these features. The normalized difference vegetation index (NDVI) image is generated from multi-spectral image, and then it is used as assistant data for building extraction.