本研究之目的為建立一套由EROS A衛星立體對影像產生正射影像及數值地表模型之程序，主要工作包含建立軌道模型、影像匹配、產生數值地表模型以及影像正射化，研究中利用此一程序可反覆迭代修正立體對影像之視差產生正射影像，並得到該測試區高解析力之數值地表模型。 在建立軌道模型部分使用載體參數修正法進行衛星軌道修正。在特徵點萃取部分使用目標導向運算元，使用一調適性門檻值萃取大量適確的特徵點以供匹配。在影像匹配中使用影像金字塔進行特徵點匹配，以減低像空間中視差大小，提高運算效率且增加匹配成功率及可靠度。在產生數值地表模型方面使用反距離權重法及區域平面密合法進行高程內插的工作並進行比較。為了避免匹配錯誤的離散點座標造成數值地表模型內插品質不佳，因此配合函數模式使用隨機模式進行過濾的工作。在影像正射化部分，本研究採用區塊逆轉換之方法，可有效提昇影像正射化之效率。 實驗結果顯示，使用此一程序，可逐步降低立體對影像的視差，整體視差標準偏差為1.5個像元平均值為0.3個像元，而產生之數值地表模型使用檢核點檢核高程均方根誤差為4公尺左右的精度。 The main purpose of this research is to build up a scheme for the generation of orthoimage and digital surface model from EROS A stereopairs. The major works include the establishment of geometric correction model, image matching, the generation of DSM and orthorectification. Utilizing this procedure one can correct the disparities of stereopairs recursively and generate orthoimages. In addition, we can derive the DSM with high resolution for the test area. On-board Data Correction Model will be applied to perform the orbit adjustment. The Target-Defined Ground Operator will be used in feature point extraction. A method for automatic determination of the adaptive threshold in the operation will be proposed, In order to increase the pull-in range and improve the reliability in image matching, image pyramid structure is considered. Two methods, i.e., Inverse Distance Weighting and Local Surface Modeling, in height interpolation will be compared in the generation of digital surface model. A filtering procedure for blunder removal will also be included in the scheme. The Orthorectification will incorporate Patch Back Projection method to accelerate the computation. Experimental results indicate that the procedure can reduce the disparities of stereopairs recursively and may reach 1.5 pixels in standard error and 0.3 pixels in mean error. The generated DSM could reach an accuracy of 4 meters in RMSE.
