Research On Inner FOV Stitching Theories And Algorithms For Sub-images Of Three Non-collinear TDI CCD Chips

The new-type TDI (Time Delay Integration) CCD chips have an inevitable tendency to replace the traditional CCD chips and beocme the mainstream sensor elements of high-resolution linear pushbroom imaging systems. Now one effective way to get a wide imaging swath is to increase the length of CCD linear array. Restricted by the number of sensor elements on each TDI CCD chip, field butting of the multiple chips is often adopted to obtain a wide FOV. However, since each TDI CCD chip is actually a CCD array in physical structure, and there is also an outer cover for each chip, it is really impossible to place the multiple TDI CCD chips directly as a straight line on the focal plane, and the non-collinear arranging style is adopted instead. Such kind of non-collinear TDI CCD cameras are important payloads onboard many commercial high-resolution imaging satellites nowadays like IKONOS, QuickBird and etc. The high resolution (HR) camera onboard CBERS-02B is also one of the typical with three TDI CCD chips arranged in a non-collinear style. Due to a non-collinear arrangement of the multiple chips on the focal plane, the control stability of the attitude angles, the undulation of the ground elevation, the variation of the integration time and so on, the sub-images separately captured by each TDI CCD chip can not directly form as an integrated image scene. Therefore, accurate and reliable inner FOV stitching which combines the sub-images to a continuous virtual image scene covering the full FOV of the camera is a fundamental step for image ground processing. However, the traditional technical routine primarily based on shift transformation of each adjacent sub-images is lack of strictness in theory and does not give full considerations to the imaging characteristics of such kind of cameras, so it is of much significance and value to do researches on theories and algorithms related to the inner FOV stitching issue of non-collinear TDI CCD cameras.This paper systematically studies the inner FOV stitching theories and algorithms of imaging data acquired by one type of non-collinear TDI CCD cameras holding three TDI CCD chips on the focal plane. The major content and innovative points are introduced as follows:(1) Imaging characteristics of the three non-collinear TDI CCD chips.Some key factors that affect the imaging quality of the three sub-images are discussed in detail, including the placement of the three TDI CCD chips on the focal plane, the control of the drift angle, the variation of the integration time, the terrain undulation and so on. Some valuable conclusions are derived out to form as the theoretical basis of inner FOV stitching algorithms developed and discussed in this paper, and also give suggestions to the design of spaceborne high-resolution imaging systems, from the perspective of high-quality data acquisition and preprocess.(2) Camera calibration model to improve the accuracy and reliability of the inner FOV stitching process.It is obvious that the imaging constraints between the adjacent TDI CCD chips give geometric foundation to the realization of seamless inner FOV stitching process. Therefore, this paper establishes the rigorous geometric sensor model of the original three sub-images, based on which the imaging geometric relation model of the adjacent TDI CCD chips are built up accordingly. By analyzing on such a relation model, it is pointed out that, to achieve the same positioning accuracy of the two adjacent sub-images which guarantees the inner FOV stitching efficiency and quality, it is an essential prerequisite to improve the measuring and calibration accuracy of both interior orientation (IO) and exterior orientation (EO) parameters of the camera. As a result, the additional polynomial coefficients based self-calibration model for single TDI CCD chip is studied, the correctness which is verified through the experiment on simulated data. Besides, with regard to the imaging characteristics of the three TDI CCD chips, the inter-relation calibration model for the adjacent TDI CCD chips by using the imaging geometry of the tie points is also introduced and discussed, experimental result indicates that the pixel coordinates prediction accuracy of the tie points is effectively improved after such kind of calibration.(3) Two improved image-space-oriented inner FOV stitching algorithms.According to the imaging characteristics of three non-collinear TDI CCD chips, the traditional mosaic algorithm by shift transformation between the adjacent sub-images is inapplicable in some cases and unable to meet the requirement of high-precision processing. In order to overcome the limitations of the existing image-space-oriented algorithm, the dissertation proposes two improved algorithms. One is based on piecewise affine transformation of the sub-image in the middle and the other is based on normalization of the integration time. The empirical results verified that both of the algorithms are simple, fast and highly reliable, which can satisfy the requirements of a sub-pixel processing level. Apparently, the image-space-oriented inner FOV stitching algorithms are all indispensable with the fast and reliable tie points extraction, since the horizontal overlap of the adjacent sub-images is quite small, to make the matching process more efficient and reliable, the imaging geometric relation of the adjacent sub-images and the prior knowledge about the horizontal and vertical offset characteristics are introduced for the prediction of pixel coordinates of the tie points, which can significantly reduce the searching scope for image matching to improve the automatic extraction efficiency of the tie points. With regard to the processing workflow, the approximate equivalent geometric sensor model of virtual image scene is established for each of the algorithms, and the high-precision RPC parameters are thus generated.(4) Two object-space-oriented inner FOV stitching algorithms respectively based on virtual CCD line and projection reference plane.Since the image-space-oriented algorithms are all dependant from the image sensor model, they are lack of theoretical strictness and the bottleneck exists when the terrain undulation and platform stability have to be considered. Relatively, the essence of the object-space-oriented inner FOV stitching is to establish the pixel coordinates relationship between the virtual image and the original image based on the geometric sensor model of the camera and the imaging constraints of the adjacent sub-images. By such a technical routine, this paper proposes two object-space-oriented inner FOV stitching algorithms, one identifies the virtual image scene as an image captured by an integrated virtual CCD line on the focal plane which is perpendicular to the moving forward direction of the camera, while the other is by rectifying the three sub-images to the given projection reference plane in object space. The principle, workflow and technical difficulties as well as the error sources are discussed for each of the algorithms. Besides, according to the practical situations that the ancillary data and the camera geometric parameters are not satisfied enough to meet the sub-pixel processing accuracy, relative orientation of the three sub-images are introduced in the workflow. The equivalent geometric sensor model of the virtual image scene is established according to the basic principle of the object-space-oriented algorithms respectively, and the high-precision RPC parameters are thus generated.(5) Verification of theories and algorithms based on simulated data and CBERS-02B HR images.The experiments are designed to verify the correctness of the theories and algorithms in this paper. Firstly, test on the offset properties of the three sub-images by simulation analysis and statistics on HR images. Then, to prove the correctness of the two object-space-oriented algorithms, experiments are carried out on simulated data when the 10 and EO parameters as well as the ground elevation information are all accurate enough, after that, simulative analysis are performed to discuss the effects of errors in attitude angle, focal length and sensor position on the prediction of pixel coordinates of the tie points. Besides, inner FOV stitching is performed on two HR images, one includes the variation of integration time while the other keep a constant integration time, the two image-space-oriented algorithms all reach the sub-pixel accuracy level, while the two object-space-oriented algorithms can also obtain the same accuracy by relative orientation of the adjacent sub-images. Finally, experiments on simulated data proves the correctness of additional polynomial coefficients based self-calibration model; selecting HR images captured on different days of a period and taking the calibrated EO parameters of the camera during such a period as the precise value, the relative position relations of the adjacent TDI CCD chips are calibrated by using the tie points. It is demonstrated that the relative positioning accuracy of the three sub-images can be improved to some degree.The proposed algorithms have been validated by non-collinear TDI CCD imaging data of some domestic high-resolution satellites launched in recently years.