Study On Filtering Of Airborne Laser Scanning Points Cloud

Airborne light detection and ranging (hereinafter referred to as LiDAR) is becoming a new method of spatial data acquisition, recording three-dimensional coordinates and reflected intensity of laser foot points by combining laser scanning, high precision dynamic GPS differential positioning technology and inertial measurement unit. As an active remote sensing, LiDAR can acquire three-dimensional geospatial information rapidly, with less susceptible to weather, season and time. Digital elevation model (shorted as DEM) plays an importance role on geo-analysis, and LiDAR provides a new quick access to a wide range of high-precision DEM.Filtering algorithms and accuracy assessment are introduced and summarized, and more attention is paid to methods used in quantitative analysis and evaluation index. Several problems influencing the filtering accuracy are discussed, for example, choosing the first or last return for ground filtering when multiple LiDAR return’s are available, the relationship between density of cloud points and pixel size of raster, and definition of DEM. The last return pulse can utilize the penetration of laser, which can better represent the ground surface. Appropriate point density can beneficial to improve the filtering accuracy, otherwise exorbitant point density will cause data redundancy and burden calculated amount while making little contribution to accuracy. Filtering DEM can be created for several filtering methods, such as morphological filtering. The Filtering DEM is designed for filtering process rather than generating DEM by using some simple padding or interpolation, so it’s not the optimal. The optimal strategy is that validated better-preformed interpolation (inverse distance weighted interpolation or Kriging interpolation) should be used to ground points judged by filtering DEM.Vulnerability scanning will be usually existent for several reasons, such as field of view is narrower than traditional photogrammetry, low flight height, Shadow effect caused by building and data missing caused by water-absorbing. Many filtering algorithms are based on the format of raster, hence, the irregular cloud data should be resampled to raster. The procedures of rasterizing are introduced systematically, and a method which fills aggregated missing data with terrain of boundary is presented based on analyzing lowest method and nearest method. Experiments show that the proposed method adapts to aggregated missing data caused by different reasons to effectively ensure the continuity of the elevation, and improves the filtering accuracy. Bridges are transportation junctions between separated ground surface, and they have many similar characteristics with ground surface. Bridges, important features, are removed by some filtering algorithms, and retained for others. To improve the reliability of bare earth detection and meet the requirement of different applications, the bridge should be detected, then should be removed or retained in the bare earth if required (as in application such as route planning). Through studying on morphological filter and regional growing, a method of bridges extracting is proposed based on difference between above filtering methods. Combining the characteristics of bridges, the bridges can be extracted from the difference of filtered DEM. The proposed method is adaptable to different bridge designs. Therefore, bridges need not parallel edge or uniform width. Furthermore, a bridge can split into parts, a bridge can be part of the whole bridge, etc.Several improvements are applied to multiscale mathematical morphological filtering: height thresholds are depended on the possible lowest features on the different structural elements. Morphological characteristics of removed features, such as area and width, are took into account with structural elements. According to the phenomenon that morphological filter erodes terrain excessively and the demand of seed ground points for region growing, a two-stage "rough-refined" filtering strategy is proposed. First, rough digital elevation model (DEM) is obtained by applying multi-scale morphological filter to airborne LiDAR data, which is called rough filter. Then, region growing is applied to gain refined DEM based on the seed ground points derived from the rough DEM. The proposed algorithm only need two parameters, max window size and height threshold for growing, which can be depend on data sets easily. The filtering accuracy of proposed algorithm is not vulnerable to window size, and superior to morphological algorithm. When tested against the ISPRS LiDAR reference datasets, the mean total error reaches3.51%(median,2.91%) and the mean Kappa score reaches87.70%(median,91.39%) with all samples. Experiment shows that proposed algorithm is highly adaptable to various landscapes for reserving the detail of terrain effectively, and also has better robustness.