| Triangulation is one of the most important steps in photogrammetry. During more than100years, it is also a focusing problem in different development periods of photogrammetry. Until the beginning of this century, Triangulation is still one of research focuses that photogrammetry researchers are concerning. In the studying history process of aerial triangulation, it mainly solves the high precision solution of exterior orientation elements including the bundle block adjustment with additional parameters, the combined adjustment with GPS/IMU and other auxiliary data, high efficiency solution of large scale sparse matrix method and the method of reliability theory such as automatic measurement error evaluation and gross error detecting.With the development of computer and related technology fields, photogrammetry enters the stage of digital photogrammetry and aerial triangulation also entered a full digital automatic aerial triangulation stage. It researches the full automatic and high precision measurement technique of the photo connection point in which Professor Ackermann proposed the least squares image matching method that can effectively improve the single point measurement accuracy of automatic image matching.With the application of GPS/IMU technology in photography, it is a feasible solution of the automatic aerial triangulation to combine multi-image matching guided by GPS/IMU and real-time bundle block adjustment. This solution breaks through the old idea of traditional automatic aerial triangulation measurement that measurement is done before the adjustment. Otherwise, it is still a challenge to realize automatic aerial triangulation without GPS/IMU and other auxiliary data for the images in photography area. In addition, the method of automatic control point measurement and automatic control point layout combining the inside and outside the industry are still the research directions of full automatic aerial triangulation automatic.With the wide application of crude photographic platform such as low-altitude UAVs in photogrammetry, the photographic mode of traditional photogrammetry is broken. Due to the condition limitation, the traditional photography condition and strict requirements of photography often cannot be fully satisfied. Some Irregular situations appear such as no GPS/IMU and other auxiliary data, the large rotation angle and large angle in the image of internal strip, and image overlap between strips beyond the requirements. Because of the appearance of the abnormal photography, new research problems are created in the already relatively mature automatic aerial triangulation, namely how to intelligently realize high precision automatic measurement of photograph connection point with the complex regional image relationship, and regional bundle adjustment. The significance of this paper is to propose the concept of intelligent aerial triangulation that hopes to achieve the combination of the brain and the computer, make full use the fast computation and fast memory of computer, and promote computer’s intelligence and guide people to complete more complex tasks under the control of the human brain, such as aerial triangulation problem in water area, the desert area, difficult and complex image relationship.In order to solve the high precision automatic measurement problem of image connection point under the complex image relationship, this paper proposed the intelligent aerial triangulation concept based theory and method that includes high precision automatic image point measurement with multi-view stereo model as the basic unit, error elimination and the refinement of the intelligent estimation of real-time imaging geometric relationship. And through the connection and merger of the multi-view stereo model, the whole region bundle adjustment can be completed. Exterior orientation elements of whole area images, the ground coordinates of the connection points and related correction parameters can be obtained. At the same time, the accuracy assessment and the weak area analysis of the whole area are done and follow-up treatment recommendations and control point layout scheme are provided in the form of the adjustment report. So this paper researches on fast multi-view stereo model based on GPS/IMU, fast multi-view stereo model based on feature image matching, multiple least squares image method based on multi-view stereo model, and region bundle adjustment of free network and error elimination of multi-view stereo model. The main research contents are as follows:(1) Automatical establishment of the initial topological relation between images of a full automatic multi-view stereo modelIn the traditional data process of aerial triangulation, the initial topology relationship between images is usually built manually according to the flight plan of survey area and the obtained images. Generally strip image list and strip relation list are established by the strip and expressed in the form of survey area project file. For images in a strip, the image sequence with60%overlapping is formed by the list. The initial relationship between survey area’s images of different strip is described by the image offset of adjacent strip. For the artificial method of setting the initial topological relation between images, there are two main problems:one is that it can’t express the corresponding relation of different strip images well. One is a large amount of manual work that is the main bottleneck of automation. According to the characteristics of digital image in photogrammetry, this paper first proposed to use the overlapping direction of a single digital stereo image pairs (left-right parallax direction) to determine automatically the installation position of the camera and create the coordinate relation between the image coordinate system and photogrammetry photograph plane system. At last, the lens system errors of image point observations can be properly corrected. Then through the study of computer graphics related theory and graph theory, topological relations between the initial images is established automatically based on the acquisition data of GPS/IMU in modern photography measurement. For the images without GPS/IMU aided navigation data, first realize the two-image matching using feature matching method through the research image matching technology based on feature. Then calculate geometrical relationship between adjacent images by searching the image of the8neighborhood. At last complete the automatic establishment of initial relationship between images, so as to guide and construct multi-image matching under the multi-view stereo model and realize automatic measurement of connection point and the analysis of control point position.(2) Study on the automatic intelligent elimination method of error matching points and gross error pointsError matching points and coarse points are almost inevitable in image matching. The important aspect of this paper is how to effectively use the robust geometric model estimation method to automatically eliminate false matching points. Here are two kinds of classic gross error elimination theory for removing the error matching points.RANSAC random sampling estimation model is put forward for removing control points with gross error in space resection. According to the basic principles of photogrammetry that image matching points between two photos should meet the coplanar condition, parameters of the coplanar condition can be calculated based on the principle of random sampling of the RANSAC. And the residual of the corresponding points is analyzed and error matching points can be automatically eliminated.For the images of the flat region, the affine transformation model is used to automatically eliminate the error matching points based on the correspondence relation of image corresponding point established by the RANSAC algorithm. For the error matching points produced in multi-view matching of multi-view stereo model. DataSnoop gross error elimination theory is used, and iterative process of weight functions and coplanar condition equations are used to establish free network bundle adjustment of small area based on multi-view stereo model. Study on the method of the error matching point elimination. Then the coplanarity condition equation and least squares adjustment principle are used to further study error detection and localization algorithm.(3) The high precision image point measurement and partition bundle adjustment of free networks and overall adjustmentHigh precision image point measurement is the key to improve the accuracy of region bundle adjustment. Otherwise, the precise elements of exterior orientation can improve the accuracy of the image matching and the reliability of the least squares image matching. Therefore, it is the key technique of intelligent aerial triangulation to research the high precision image matching and region bundle adjustment.This paper adopts multi-view stereo model as a minimal region of high precision image point measurement and free network bundle adjustment. Pyramid images are used for channel matching from coarse to fine scale based on feature point. And according to the hierarchical matching results, free network bundle adjustment can be completed. While gross error is eliminated, accurate estimation of each orientation element image is completed. Finally, in the original image, the high precision image point measurement and partition bundle adjustment of free network can be completed by using accurate estimation of orientation elements and multiple image least squares adjustment based on collinear condition.The matching results of multi-view stereo model will be stored by image. Weak zone is checked based on the corresponding point distribution and the number of corresponding point observation is checked according to adjustment minimizing condition.(4) Control point layout scheme and automatic control point measurementAccording to the error theory, the weak area of bundle block adjustment is analyzed. In addition to the general principle of control point layout and the existing control point layout standard, the relationship between regional images is also considered to automatically predict the position of control points in increasing control points of weak area.Feature points are extracted according to the control point prediction and the corresponding image points are found. The multi-image matching method based on epipolar constraint is used and the matching points are processed by gross error detection.Control points are output into small images and text, and displayed in the mobile equipment (such as notebook computer, PDA, IPad and intelligent mobile phone etc) in order to guide the field workers measure the control point coordinate at the specified location. |
PDF Full Text Request