|
Sensor Integration and Image Georeferencing in Support of Airborne Remote Sensing Applications 2. Image Direct Georeferencing From the discussion in the previous section, it is clear that the indirect georeferencing of airborne remote sensing images has some operational as well as practical problems. Therefore, there has been a major requirement by the mobile mapping community to eliminate or at least reduce the required number of GCPs. This can be achieved by installing navigation sensors, beside the imaging sensors, on board the aircraft to determine the EOPs of the imaging sensor directly. With the advent of the Global Positioning System (GPS), the idea of installing a GPS to determine the position of the imaging sensor Perspective Center (PC) at each exposure was performed and used for some time. However, this could not eliminate the need for GCPs but only reduced the required number of them, see Hofmann-Wellenhof et al. (1998) for more details. As part of the intensive research of multi-sensor integration at the University of Calgary, a major part was directed towards the development of an integrated navigation system that integrates a Differential GPS (DGPS) and an Inertial Navigation System (INS) for accurate position and attitude determination. In such a system, the GPS is used to provide positions while the INS is used to provide orientations. By installing this INS/GPS integrated system with the imaging sensor, the EOPs can be determined without any GCPs. This approach was first introduced by University of Calgary researches and is named direct georeferencing, see for example: Schwarz et al. (1984), Schwarz et al. (1993) and Schwarz (1995). In INS/GPS applications, the initial trajectory (velocity, position and attitude) is obtained by integrating the output of the inertial sensors (accelerometer specific forces and gyro angular rates). This is performed through the INS mechanization (navigation) equations, which are in fact, a set of non-linear differential equations. The solution of these equations is compared to the provided GPS solution and the differences are used to estimate and compensate for the INS errors through a Kalman Filter (KF), see Figure 1, where r, v, att and d represent the position, velocity, attitude and estimated errors, respectively.  Fig.1 INS/GPS Navigation Architecture in Direct Georeferencing Applications In direct georeferencing by INS/DGPS integration, three coordinate frames are dealt with: the b-frame with its center at the Inertial Measuring Unit (IMU) center, the imaging sensor frame (c-frame) that has its center at the PC of the imaging sensor and the mapping frame (m-frame) where the final object coordinates are required and is usually considered as the local-level frame (l-frame). The direct georeferencing model is given in El-Sheimy (1996) as:  where rjm is the position vector of an object j; rGPSm is the position vector of the GPS antenna interpolated to the time of exposure t; sj is a scale factor for an object per image that relates image coordinates to the object coordinates, which is usually obtained implicitly using image stereopair processing techniques, laser scanner or a Digital Terrain Model (DTM), El-Sheimy (1996); Rcb is the rotation matrix between the c-frame and b-frame (assumed to be constant for the same installation and is determined by calibration before or during the mission); rjc is the vector of image coordinates given in the c-frame; acb and aGPSb are constant vectors between the IMU center and both the imaging sensor PC and the GPS antenna center given in the b-frame (determined during calibration process prior to mission by traditional surveying, i.e. total station). In the above model, Rbm is obtained from the solution of the INS mechanization equations. The graphical representation of the direct georeferencing model is illustrated in Figure 2.  Fig.2 Airborne Remote Sensing Image Georeferencing Components Following the introduction of the direct georeferencing approach, it has been later widely used in many airborne remote sensing applications, either for research or for commercial production. In the following paragraphs, results obtained using INS/GPS for direct georeferencing will be summarized. The accuracies for the different systems are the Root Mean Square (RMS) values of the differences between the INS/GPS/imaging solution and the reference solution that is provided by well-known established GCPs. In airborne mapping applications, the obtained accuracy using INS/GPS/imaging sensor configuration depends mainly on the scale of photography (i.e. the flying height). Using frame-based aerial cameras, the reported accuracies are 10-20 cm for easting and northing and 8-32 arcsec for attitude angles (roll, pitch and azimuth). The corresponding height accuracy is 10-30 cm, for more details see Skaloud (1995); Abdullah (1997); Hutton et al. (1997); Reid and Lithopoulos (1998); Reid et al. (1998); Skaloud (1999); Cramer et al. (2000). In case of CCD digital cameras, the accuracy for airborne applications also depends on the camera resolution. The results given in Grejner-Brzezinska and Toth (1998) using a high-resolution 4k*4k CCD camera showed positional accuracies of 19, 20 and 32 cm in X, Y and Z directions, respectively. Using dual (nadir and oblique) CCD cameras, Mostafa and Schwarz (1999) reported accuracies of 54, 61 and 78 cm in X, Y and Z coordinates using a single stereopair of a nadir and oblique images. With the same system of dual cameras, Mostafa (1999) showed after using a 3*3 block of nadir and oblique images corresponding accuracies of 22, 24, and 34 cm, respectively, can be achieved. Laser scanners, which are mainly used for generating DTMs & Digital Elevation Models (DEMs) and for mapping forests, vegetation and urban areas. The reported accuracies are in the range of 20-60 cm, 20-60 cm and 10-25 cm for easting, northing and height, respectively. See for instance Kimura et al. (1999); Baltsavias (1999); Mohamed et al. (2001) and Maas (2003). Pushbroom linear scanners are used in applications that require an accuracy of 2-10m (Alamús and Talaya, 2000). This was confirmed by Cosandier (1999) who obtained accuracies of 2.5m - 3.5m for each channel with the Compact Airborne Spectrographic Imager (casi) system. With InterFerometric SAR (IFSAR) systems, their main usage is the determination of DEMs, especially in areas with heavy vegetation. Arbiol and González (2000) showed planimetric accuracy of 8.7m and vertical accuracy of 5.7m. Specifications given for the DEMs generated by the Intermap Technologies Ltd. STAR-3i airborne system confirmed obtained vertical accuracies in the order of 0.5-3m (post spacing of 5m) with a corresponding horizonta1 accuracy of 2.00m on slopes less than 20° (Intermap Technologies, 2005). 3. Direct Georeferencing Accuracy The final image direct georeferencing accuracy obtained from an INS/GPS navigation system (i.e. the navigation accuracy at the imaging sensor PC), regardless of the imaging sensor type, accuracy or quality, is a function of the complete INS and GPS processing chain (error control implementation and KF design), IMU accuracy, GPS receiver quality, the alignment between the imaging and navigation sensors (see Equation 1) as well as the airborne data collection circumstances and environment. Hence, it is depending on both the measurement and processing stages. This involves the INS/GPS system estimated position and orientation angles, the INS/GPS/Imaging sensors alignment and time synchronization, the airborne operation circumstances and the sensors’ noise characteristics. Therefore, to improve the image direct georeferencing accuracy obtained from the INS/GPS integrated system, a number of factors have to be considered:
|