The topic of this research is the development of a mathematical model for the georeferencing of imagery acquired by multi-line CCD array sensors, carried on air- or spacecraft. Linear array sensors are digital optical cameras widely used for the acquisition of panchromatic and multispectral images in pushbroom mode with spatial resolution ranging from few centimeters (airborne sensors) up to hundreds meters (spaceborne sensors). The images have very high potentials for photogrammetric mapping at different scales and for remote sensing applications. For example, they can be used for the generation of Digital Elevation Models (DEM), that represent an important basis for the creation of Geographic Information Systems, and the production of 3D texture models for visualization and animation purposes.
In the classical photogrammetric chain that starts from the radiometric preprocessing of the raw images and goes to the generation of products like the DEMs, the orientation of the images is a fundamental step and its accuracy is a crucial issue during the evaluation of the entire system. For pushbroom sensors the triangulation and photogrammetric point determination are rather different compared to the standard approaches for full frame imagery and require special investigations on the sensor geometry and the acquisition mode.
Today various models based on different approaches have been developed, but few of them are rigorous and can be used for a wide class of pushbroom sensors. In general a rigorous sensor model aims to describe the relationship between image and ground coordinates, according to the physical properties of the image acquisition. The functional model is based on the collinearity equations. The sensor model presented in this thesis had to fulfill the requirement of being rigorous and at the same time as flexible as possible and adaptable to a wide class of linear array sensors. In fact pushbroom scanners in use show different geometric characteristics (optical systems, number ofCCD lines, scanning mode, stereoscopy) and for each data set specific information are available (ephemeris, GPS/INS observations, calibration, other internal parameters). Therefore the model needs to be dependent on a certain number of parameters that may change for each sensor.
According to the availability of information on the sensor internal and external orientation, the proposed model includes two different orientation approaches.
The first one, the direct georeferencing one, is based on the estimations of the ground coordinates of the points measured in the images through a forward intersection, using the external orientation provided by GPS and INS instruments or interpolated by ephemeris or computed using the orbital parameters (satellite case). This approach does not require any ground control points, except for final checking, and does not estimate any additional parameters for the correction of the interior and exterior orientation. For this reason, the accuracy of this method depends on the accuracy of the external and internal orientation data.
The alternative orientation method, based on indirect georeferencing, is used if the sensor external and internal orientation is not available or not enough accurate for high-precision photogrammetric mapping. This approach is a self-calibrating bundle adjustment. The sensor position and attitude are modelled with 2nd order piecewise polynomial functions (PPM) depending on time. Constraints on the segment borders assure the continuity of the functions, together with their first and second derivatives. Using pseudo-observations on the PPM parameters, the polynomial degree can be reduced to one (linear functions) or even to zero (constant functions). If GPS and INS are available, they are integrated in the PPM. For the self-calibration, additional parameters (APs) are used to model the lens internal parameters and distortions and the linear arrays displacements in the focal plane.
The parameters modelling the internal and external orientation, together with the ground coordinates of tie and control points, are estimated through a least-squares bundle adjustment using well distributed ground control points. The use of pseudo-observations allows the user to run the adjustment fixing any unknown parameters to certain values. This option is very useful not only for the external orientation modelling, but also for the analysis of the single self-calibration parameter's influence. The weights for the observations and pseudo-observations are determined according to the measurement accuracy. A blunder detection procedure is integrate for the automatic detection of wrong image coordinate measurement. The adjustment results are analyzed in terms of internal and external accuracy. The APs to be estimated are chosen according to their correlations with the other unknown parameters (ground coordinates of tie points and PPM parameters). A software has been developed under Unix environment in C language.
The flexibility of the model has been proved by testing it on MOMS-P2, SPOT-5/HRS, ASTER, MISR and EROS-Al stereo images. These sensors have different characteristics (single-lens and multi-lens optical systems, various number of linear arrays, synchronous and asynchronous acquisition modes), covering a wide range of possible acquisition geometries.
For each dataset both the direct and indirect models have been used and in all cases the direct georeferencing was not accurate enough for high accurate mapping. The indirect model has been applied with different ground control points distributions (when possible), varying the PPM configurations (number of segments, polynomials degree) and with and without self-calibration. Excluding EROS-Al, all the imagery has been oriented with sub-pixels accuracy in the check points using a minimum of 6 ground control points. In case of EROS-Al, an accuracy in the range of 1 to 2 pixels has been achieved, due the lack of information on the geometry of the sensor asynchronous acquisition. For the ASTER and SPOT-5/HRS datasets, a DEM has also been generated and compared to some reference DEMs.
New cameras can be easily integrated in the model, because the required sensor information are accessible in literature as well as in the web. If no information on the sensor internal orientation is available, the model supposes that the CCD lines are parallel to each other in the focal plane and perpendicular to the flight direction and estimates any systematic error through the self-calibration. The satellite's position and velocity vectors, usually contained in the ephemeris, are required in order to compute the initial approximations for the PPM parameters. If this information is not available, the Keplerian elements can be used to estimate the nominal trajectory. For pushbroom scanners carried on airplane or helicopter the GPS and INS measurements are indispensable, due to the un-predictability of the trajectory.