Reconstruction Of 3D Points And Curve Structures Based On Multiple Images

In recent years, with digital camera being used widely, its resolution and imaging quality are enhanced greatly. The price is becoming lower and lower. In addition, the progress in image processing and pattern recognition is increasingly fast. The rapid, flexible and low cost 3D data acquisition techniques by using digital cameras have become the furthest promising 3D measurement ways. In this thesis, 3D point and curve structure reconstruction from multiple images obtained by one digital camera is deeply studied based on the rigorous photogrammetry theories and the latest research achievements in computer vision. A new method of flexible 3D measurement is proposed for industrial products.To ensure reliability of feature recognition and to heighten the precision of image processing and the level of measurement automatization, a set of reference points (including coded points and uncoded points) and marked object curves are applied. The coded points are used as an assistant utility, whereas the uncoded points and the marked curve structures are the elements to be reconstructed. The main contents and contributions of this thesis are as follows:(1) An algorithm of automatic reference point detection is put forward and implemented. Firstly, the targets are extracted from images according to their size, shape and intensity etc. Then an improved method to identify the unique identity of every coded target is put forward. At last, the multiple-threshold-line algorithm is applied to get the subpixel location of both coded and uncoded targets. Practical examples show that the results are satisfying.(2) A non-iterative image feature matching algorithm based on reference point correspondences is proposed. The initial matching of the uncoded target correspondences between an image pair is established according to the similarity and compatibility, which are based on the ID correspondences of the coded targets. The outliers in the initial matching of the uncoded target are eliminated according to three rules to finally obtain the uncoded target correspondences. Practical examples show that the algorithm is rapid, robust and is of high precision and matching ratio.(3) A new algorithm for recovery of camera poses and 3D reconstruction of reference points is presented and realized. A good initial value is estimated for the projective reconstrucrion by the matrix factorization method using coded targets as original input, which leads to accurate estimation of the camera motion matrices and. The coordinates of the uncoded targets in 3D projective space is then recovered by using the stereo triangulation. The coordinates of the uncoded targets and camera motion matrices are further optimized by using the bundle adjustment. Finally, the projective reconstruction is upgraded into Euclidean one by making use of the homography. In addition, the robustness is remarkably increased since the essential condition rank( ? )=3 of the absolute quadric is imposed to the homography. Compared to the existing method, the reprojection error is greatly reduced by the proposed one. The proposed method has been applied in multi-view point cloud registration.(4) A novel method for measurement of 3D curve structures by using one hand-held digital camera is proposed. Firstly, combining convenient interaction with the energy optimization method, object curves are extracted semi-automatically. Secondly, a list of points on the object curve is reconstructed by using stereo triangulation based on optimization of the object curve correspondences. Finally, the measurement of object curves and reconstruction of models for multiple examples are carried out by using this algorithm. The reconstruction error was analyzed to demonstrate that the method is an effective technique.(5) A prototype system for 3D photography measurement for industrial products is developed. The reconstruction examples of industrial products demonstrate that the presented methods are effective.The proposed method has the following advantages: firstly, the hardware used for the measurement is quite simple; secondly, the measuring process is very convenient and flexible; thirdly, no registration is needed for the digitized data obtained from different views; and finally there is no redundancy in the digitized data for model reconstruction. The method can be widely used in reverse engineering and quality inspection.