| Summary: | 碩士 === 國立成功大學 === 測量及空間資訊學系碩博士班 === 98 === Airborne LiDAR has become the primary technology of DEM generation. In order to produce DEM from airborne LiDAR data, the major process is filtering non-ground points. Mane filtering methods have been developed. All filters are designed to keep as many correct ground points as possible. However, there is no perfect filter yet. Every filter more or less misclassifies some non ground points to ground point, or the other way round. Owing to the widespread application of DEM, how to assess the quality of a DEM product is an important topic. For this reason, the paper investigates the quality influences of process of DEM generation from airborne LiDAR data.
The quality of DEM generated from airborne LiDAR data is conventionally assessed with the classification accuracy of point cloud filtering, or checking the elevation differences of the sampling points. This paper focuses on the study of quality control of DEM generation process. Three methods applied to assess DEM quality were investigated using ISPRS LiDAR test datasets. The first method is the evaluation of filtering accuracy. The filtered results are evaluated using the error matrix and the corresponding statistics indices. The second method is elevation accuracy. The DEM difference between a generated DEM and the reference dataset is computed for the evaluation. The third method is to locate large errors using the normalized DEM differences in which DEM differences are normalized using terrain gradients. The combination of the three methods is suggested for the quality control of DEM generation from airborne LiDAR data.
The experiments conducted in this paper mainly investigate three issues about the quality control of DEM generation from airborne LiDAR data. The first issue is the suitable of quality assessment methods of DEM generation. The experimented results suggest that one can not solely rely on statistics indices of point cloud classification to assess DEM quality. The RMSE are histogram of DEM difference should be investigated as well. The second issue is that the performance of a point cloud filter may vary subject to the terrain types. After all, none of filter methods has been claimed to be perfect. Manual editing is, therefore, suggested to improve DEM quality. Finally, how manual editing may improve the quality of DEM generation is studied. In accordance with the demand of DEM quality, a producer need to take the time and labor consumption into account.
|
|---|
