本研究之目的為針對EROS A衛星影像的幾何特性建立幾何改正模式，主要工作包含軌道修正及影像正射化兩個部份。研究中將使用衛星載體資料並配合地面控制點進行軌道修正，應用修正後之軌道配合數值地形模型進行影像正射化。 在軌道修正方面，分別建立光束法及載體參數修正法兩種模式進行衛星軌道修正。於光束法平差模式中，以時間函數描述衛星位置及姿態，使用加權之附加觀測方程式對外方位進行約制，進行方位求解，並以最小二乘過濾模式對軌道進行過濾修正。於載體參數修正模式中，由於衛星位置及姿態的高相關性，將固定衛星姿態修正軌道之位置，研究中使用衛星載體資料所提供之軌道為初始值，將衛星軌道修正量視為時間之低階多項式，使用地面控制點進行修正，在地面控制點足夠時，使用最小二乘過濾模式對軌道進行過濾修正。 在影像正射化方面，使用逆轉換模式建立物空間及像空間之對應關係，為提升逆轉換模式之運算效率，使用網格化方式將影像分成區塊進行處理。 實驗結果顯示，就EROS A之單張像幅而言，軌道修正時檢核點之均方根誤差可達2至2.5個像元之精度。影像正射化，以逆轉換模式所產生之正射影像，其檢核點均方根誤差在像平面上能達2至2.5個像元之精度。 The main purpose of this research is to build up the geometric correction model for EROS A satellite. The major works include orbit adjustment and orthorectification. The orbit modeling will be done by using the onboard data and GCPs (Ground Control Points); while the orthorectification will be done by using the precise orbit and DTM (Digital Terrain Model). Both Bundle Adjustment and Onboard Data Correction Model will be applied to do the orbit adjustment. In the Bundle Adjustment, we assume the exterior parameters as polynomial functions. To stabilize the system, we have added weight matrix for the parameters estimates; and to fine tune the orbit, we use the Least Squares Filtering technique. In the Onboard Data Correction Model, the high correlation between the satellite position and satellite attitude, we fix the satellite attitude to correct the satellite position, the orientation parameters is initialized from the onboard data. We perform the correction for on-board parameters as polynomial functions. Then, by using a small number of ground control points, we adjust the satellite’s orbit accurately. Finally, we use the Least Squares Filtering technique to fine tune the orbit. In orthorectification, we use the indirect method to generate the orthoimage. The indirect method projects the 3D object point on to 2D image space. In order to accelerate the computation, we divide the image into a number of tiles in the orthographic correction. Experimental results indicate that the proposed scheme may reach an accuracy of better than two pixels in the image scale for an image sampled with an asynchronous ratio of 13.
