Approach To The Key Issues Of 3D Model Pipeline Based On Implicit Representation

As important research contents of the National Natural Science Foundation of China (Grant No. 60275012), the Natural Science Research Grant of Higher Education of Guangdong province (Grant No. 04Z010), and the Science & Technology Bureau of Shenzhen (Grant No. 200341), this thesis focuses on the three key issues of three dimensional pipeline-modeling based on implicit representation. Those issues involve in the calibration method for three dimensional vision system, the registration method for multi-view range images and the integration method for multi-view range images. Both theoretical and experimental aspects are considered, and some innovative research outcomes have been accomplished.The main research contents and contributions made by this thesis are summarized as follows:1. The surface reconstruction method using variational implicit surface with radial basis function interpolation was exploited in detail. On the basis of this analysis, we developed an efficient scheme for the registration and the integration of multiple range images. Moreover, we built up an in-house 3D scanner with which we demonstrated our approaches.2. With the concept of benchmark transmission and shifted point array encoding we developed a method for calibrating our 3D scanner. Experiment results showed that after the proposed calibration procedure the 3-D vision system was able to achieve such an accuracy as the standard deviation in X direction was 0.288mm, 0.238mm in Y direction, and 0.292mm in Z direction, respectively, for a measuring volume of 300×300×80mm~3.3. To register multiple range images, we presented a method for the pose estimation for 3D scanner based on the principle of fringe projection. The projector could be conceptually regarded as a camera that works in a reversed mode. To determine homologous points in the proposed camera/projector configuration, the phase map was converted to the u, v coordinates using two orthogonal fringe sequences. Taking epipolar geometry into account, this approach could further realize the estimation of arbitrary pose of view point and achieve a self-calibration of the 3D scanner by minimizing the squared distances between those imaging points and their corresponding epipolar lines. With this procedure, we were able to simultaneously calculate the parameters of the orientation and the system structure. Parameters with moderate accuracy could be estimated even if the input data deteriorated with the noise was presented. Because this procedure didn't need a priori 3D calibration reference, the practicability of the 3D scanner had been therefore improved.4. The signed-distance function has been used frequently in most of integration methods based on implicit representation. In this thesis, we expanded the definition of signed-distance function in a vector space. Furthermore we developed a generalized implicit equation (GIE) by virtue of a strict mathematical formulation and a geometric reasoning. The GIE was a canonical implicit equation and was constructed by using least-mean-square approximation and vector-map reasoning. With the GIE, we could reexplain how to establish a weighted function and other geometric relations. Finally, we integrated multiple range images with the technique of variational implicit surfaces. In this process, the embedded relationship among the neighboring viewpoints and reliability of data points used in the weighted function should be considered.