Research On Correspondence Of 3D Data

In recent years,with the rapid development of 3D model scanning hardware and the maturity of multi-view 3D model reconstruction technology,3D dynamic sequences are increasing rapidly,and are widely used in various fields.Therefore,the 3D dynamic data corresponding technology has become a hot issue in current research.The thesis focuses on the research of 3D dynamic surface correspondence under different attitudes.The main research work includes:1.Existing corresponding algorithms will cause error correspondence since geometric signatures can't remain stable and highly similar under different poses.This paper is focusing on corresponding 3D dynamic surface based on space-time constraints.Firstly,the algorithm constructs the energy optimization function according to the non-rigid deformation model.Secondly,sparse correspondence is computed by optimizing the energy function.Finally,the algorithm of surface sampling and isometric correspondence is used to solve the dense matching problem.In experimental part,the analysis and quantification of different 3D motions are carried out,and it turns out that our algorithm can improve the correspondence accuracy.2.Isometric correspondence is an important technique for surface correspondence.Recently,numerous algorithms have been proposed to build isometric correspondence.However,those methods tend to produce error-prone corresponding due to topological variation and noises of dynamic surface.To address this issue,we propose a dynamic surface correspondence method by computing maximum spatial-temporal isometric cluster.Firstly,the algorithm defines a maximum isometric cluster score to measure the correspondence quality of each cluster in product space.Then,the maximum problem is formulated into a quadratic programming problem.Furthermore,we define a similarity function which encodes spatial-temporal consistence of dynamic surface explicitly.It can greatly reduce the solving dimension,and improve correspondence accuracy.Finally,the result is extended to dense correspondence by geodesic distance vector.Experimental results show that our algorithm can generate consistent correspondence on three databases of surface sequences,which outperforms existing state-of-the-art algorithms.3.Random walk is widely used in optimization problems because of its fast convergence.In this paper,a space-time isometric random walk graph is proposed for alignment of 3D dynamic surface.The algorithm defines graph nodes according to the product space of two adjacent frames.In addition,it reduces nodes based on spatial-temporal adjacency.Edge weights of the constructed graph are defined by geodesic distance.In this way,the alignment problem can be solved by random walk.Finally,the alignment results are computed by Markov chain theory.The experimental results of different dynamic surface databases show that the proposed algorithm can obtain a consistent alignment for three-dimensional dynamic surface with obvious noise and holes.Furthermore,its performance and effiency are better than the existing algorithms.