DEM EXTRACTION FROM WORLDVIEW-3 STEREO-IMAGES AND ACCURACY EVALUATION
Keywords: Worldview-3, Stereo-images, DEM Extraction, Accuracy Evaluation, LiDAR Point Cloud
Abstract. This paper validates the potentials of Worldview-3 satellite images in large scale topographic mapping, by choosing Worldview-3 along-track stereo-images of Yi Mountain area in Shandong province China for DEM extraction and accuracy evaluation. Firstly, eighteen accurate and evenly-distributed GPS points are collected in field and used as GCPs/check points, the image points of which are accurately measured, and also tie points are extracted from image matching; then, the RFM-based block adjustment to compensate the systematic error in image orientation is carried out and the geo-positioning accuracy is calculated and analysed; next, for the two stereo-pairs of the block, DSMs are separately constructed and mosaicked as an entirety, and also the corresponding DEM is subsequently generated; finally, compared with the selected check points from high-precision airborne LiDAR point cloud covering the same test area, the accuracy of the generated DEM with 2-meter grid spacing is evaluated by the maximum (max.), minimum (min.), mean and standard deviation (std.) values of elevation biases. It is demonstrated that, for Worldview-3 stereo-images used in our research, the planimetric accuracy without GCPs is about 2.16 m (mean error) and 0.55 (std. error), which is superior to the nominal value, while the vertical accuracy is about -1.61 m (mean error) and 0.49 m (std. error); with a small amount of GCPs located in the center and four corners of the test area, the systematic error can be well compensated. The std. value of elevation biases between the generated DEM and the 7256 LiDAR check points are about 0.62 m. If considering the potential uncertainties in the image point measurement, stereo matching and also elevation editing, the accuracy of generating DEM from Worldview-3 stereo-images should be more desirable. Judging from the results, Worldview-3 has the potential for 1:5000 or even larger scale mapping application.