The Study Of Strips Reconstruction Oriented Data Processing Methodology In Low-altitude Photogrammetry

With the development of aviation, computer, information, sensor technology, low-altitude digital photogrammetry has become one of the important technical means of earth observation with flexible, fast, efficient, and low cost. Low-altitude photography is suitable for the data acquisition of multi-source, multi-scale, high-resolution, and the data can intuitively, comprehensively and dynamically reflect land surface, so low-altitude photogrammetry can make up some shortcomings of traditional satellite and aerial photogrammetry. The UAV, airship for low-altitude photogrammetry has been widely used in the fields of resource and environment investigation, urban planning, disaster monitoring.Compared with the traditional aerial or aerospace photography, it is very different in low-altitude photography platform, flight altitude, sensor, photographic pose, and the platform is easily affected by air flow. The acquired images maybe small coverage area, large amount, irregular overlap, big rotation angle etc. Fast response is a big advantage of low-altitude photogrammetry, so automatic, fast image processing is of importance. The traditional data processing method of aerial photogrammetry is unsuitable for low-altitude photogrammetry, should be more reliable, efficient, automatic.The paper aims to improve robustness, automation, and efficiency of low-altitude image processing, these key technologies include strips automatic reconstruction, non-metric camera calibration, image matching etc. The main contents of this dissertation include:(1) Research on strips automatic reconstruction method without manual intervention or external information supported.It is low-level automation or universality for reconstructing strips by manual intervention or POS, so a strips automatic reconstruction method without manual intervention or external information supported was proposed in this paper, which can reconstruct strips by image internal features. In the method, vocabulary tree of SIFT features is constructed by clustering analysis, SIFT features of each image are quantified to the nodes by TF-IDF numerical statistical method for transforming each image to a vector, images correlation network is constructed by feature vector similarity, image matching and graph theory, finally strips are reconstructed automatically by tracing strips trend lines and growing the adjacent images. Experimental results show that the method can rearrange the strips of low-altitude images automatically and be high efficiency.(2) Research on reference objects automatic detection based low-altitude camera automatic calibration.Most of low-altitude cameras are non-metric with big image distortion, and need to change lens for different aerial application frequently, so fast and robust camera calibration has very important practical value. In this paper, morphology and local Hough transform for reference objects automatic detection based non-metric camera calibration was proposed, the reference objects are grid points with high precision generated by operating system functions, and the method solves the parameters of interior orientation and lens distortion by2D-DLT and bundle adjustment. Experimental results show that the method can calibrate low-altitude camera for good anti-noise, strong practicality and reliability.(3) Research on low-altitude image matching strategy combined with SIFT, ASIFT and phase correlation.The common image matching methods are unsuitable for low-altitude image matching with the difference of scale, rotation angle, terrain and repeated texture, so an image matching strategy combined with SIFT, ASIFT and phase correlation was proposed. In the method, rough matching is obtained by SIFT or ASIFT or phase correlation, grid-based Harris sub-pixel corners are extracted in one of stereo images, and accurate image matching is obtained by searching the corresponding points with epipolar correlation and least square method in other stereo images. Experimental results show that the method maybe reliable and practical for low-altitude image matching, and can get enough and well-distributed corresponding points.(4) Research on low-altitude relative orientation and model connection.It is difficult to obtain the initial value of low-altitude image relative orientation with the traditional aerial image relative orientation method, so the precision and iterative efficiency of low-altitude image relative orientation solution is affected. In this paper, the two step method combined with direct solution and continuous relative orientation was proposed, which can obtain initial value by direct solution in first step and refine relative orientation by iterative solution in second step. Experimental results show that the two steps solution of relative orientation can improve the precision of relative orientation. Meanwhile, it is difficult to represent accumulated error and distorted strips by simple mathematical formula in low-altitude strips stereo model connection, to solve the problem, the accumulated error is adjusted by changing reference image and nonlinear polynomial method. Experimental results show that the method can significantly reduce the accumulated error of model connection.According to the all research contents, this paper makes out the computer program by Visual C++language including low-altitude camera calibration&image distortion correction, strips automatic reconstruction, image matching etc. In order to further prove the efficiency of the methods, finally the paper researches on low-altitude image stitching, orthophoto making and three-dimensional simulation.