| Image smear is one of the basic problems in aerial remote sensing technique. Discussion on how to accurately model image smear has very important value in both theoretical research and application for improving the performance of aerial camera imagery. From the perspective of rigid body kinematics, image smear can be summarized as two parts, i.e., camera’s 6-DOF rigid-body motion in the three-dimensional coordinate system and TDI adder’s linear motion in two-dimensional image coordinate system. The former includes the image smear caused by aircraft forward flying and camera gimbal rotation, whereas the latter includes that introduced by TDI discrete charge transfer. At present, the research on image smear modelling, most of which is based on simplified models under different constraint with limitations, still needs to improve.Several parameters, including TDI stages, exposure time and binning mode, could impact image smear. However, if we choose the minimum of image smear as the only object function of exposure algorithm, the SNR of image could decline correspondingly, which might leads to image quality degradation. Consequently, the object function of optimum exposure algorithm should take image gray level, MTF model and SNR model into account. How to build image motion model for different TDI sensors and aerial cameras? How to solve the exposure optimization problem? We discuss these issues in this dissertation, the details are as follow:First, we study the structure and the scheme of TDI-CCD, TDI-CMOS with different shutter. With the assumption of the optical system being linear filter and sensor pixel being integral sampler, models for the above sensors are developed, especially the MTF theoretical model for Multi-line CMOS. Simulation result shows that the MTF of Multi-line CMOS is superior to TDI-CCD when the exposure time is within 1/2φ row transfer period(φ refers to clock phase number).Secondly, image smear models of panoramic aerial camera and pushbroom aerial camera are developed respectively. Due to the diversity of these two types of aerial camera, different mathematical methods are used. For panoramic camera, we simplify the transform matrix and derive the Eulerian description by using kinematics and differential method. In this case, comparing to Lagrangian description, the image smear is given by functions in terms of fiexed grid, which is more accurate and intuitive. The results of computer simulation and experiment indicate that forward image motion and scan image motion could not be fully compensated with scan-mirror head and TDI sensor. For pushbroom camera, we develop the model by using difference quotient method and first order approximation. In addition, we discuss geolocation error caused by image smear. Simulation results show that when centroid is used as image location operator and middle exposure pulse used as POS event trigger, geolocation accuracy could reach maximum.Finally, from the perspective of image quality, the concept of exposure optimization is proposed. Instead of using average image DN as the only object function, MTF model and SNR model, which are in terms of TDI stages number, analog gain, exposure time and binning mode, are used to reformulate object function. Since the equalities and inequalities and not easy to solve, we use equivalent smoothing function to reconstruct the problem and solve it with Newton algorithm. Numerical experiments prove the validity of the algorithm. |
PDF Full Text Request