Comparison of 3D physical and empirical models for generating DSMs from stereo HR images.

摘要

This research study addressed and compared 3D physical and empirical models for stereo-processing and the generation of digital surface models (DSMs) from different stereo high-resolution (HR) sensors (Ikonos and QuickBird). The 3D physical model is Toutin's Model (TM) developed at the Canada Centre for Remote Sensing, and the empirical model is the rational function model (RFM). The study also evaluated the conditions of experimentation to appropriately use these 3D models. The study was conducted in Quebec City, Quebec, Canada. The first results on stereo-bundle adjustments demonstrated that TM and vendor-supplied RFMS gave similar results with Ikonos as soon as RFM was refined with a shift computed from one GCP. On the other hand, TM gave better results than vendor-supplied RFMs with QuickBird regardless of the polynomial order and the number of GCPs. Due to its relief dependency, QuickBird RFM needed to be refined at least with linear functions computed from at least 6 to 10 GCPs. Some large errors were, however, noted on forward image RFM in column. The DSMs were then generated using an intensity matching approach and compared to 0.2 m accurate lidar elevation data. Because DSMS included the height of land cover (trees, houses), elevation linear errors with 90% confidence level (LE90) were computed and compared for the entire area and 3 land cover classes (forests, urban/residential, bare surfaces). TM and vendor-supplied RFMs with Ikonos, regardless of the method and GCP number, achieved comparable results for all classes, while TM achieved overall better results than vendor-supplied RFMs with QuickBird. All results demonstrated the necessity of refining Ikonos RFM with a shift and one GCP only and QuickBird RFM with 1st-order linear functions and 6-10 GCPs due to its relief dependency..