Font Size: a A A

The Research Of Non-rigid Structure From Motion Based On Manifold Groups Of Motion

Posted on:2013-10-17Degree:MasterType:Thesis
Country:ChinaCandidate:J B GuoFull Text:PDF
GTID:2248330371986100Subject:Signal and Information Processing
Abstract/Summary:PDF Full Text Request
Three-Dimensional reconstruction from image sequences is an important andessential task of computer vision. In the real world, many scenarios are non-rigid ordynamic. Therefore, non-rigid three-dimensional motion analysis has become thepresent three-dimensional motion visual focus of the study.Existing non-rigid motion reconstruction algorithm is based on the idea thatobserved shapes can be represented as a linear combination of a compact set of basisshapes. The emergence of this idea of three-dimensional motion analysis of the visualinto a new stage of development, after development in recent years researchers havemade a lot of movement based on the idea of three-dimensional reconstruction of thetheory. Although these methods get a lot of solutions of non-rigid structure frommotion, the shape of the object must be re-estimated if the object has changed orsports scene changes. This will result in the previous algorithm does not apply andbring to problem solving difficulties.To solve the above problem this paper made the following aspects of work:(1)Find a unified model to describe the movement patterns of different forms ofnon-rigid motion. Although the existing non-rigid motion reconstruction algorithmsgenerally agree this assumption that observed shapes can be represented as a linearcombination of a compact set of basis shapes, however, the biggest drawback is thatthere cannot be for all the non-rigid motion to establish a unified form of the model.Existing algorithm of non-rigid structure from motion only for specific occasions,when occasions change we must re-estimate the shape of the base. For example, thebase shape to describe human facial movement cannot be applied to the base body towalking and dance. This paper researches the common form of all non-rigid motion and the physical nature and establishes manifold groups of motion To describe thedifferent types of non-rigid motion. Experiments show that this model can be a gooddescription of various non-rigid body movement patterns.(2) We get a solution of an improved non-rigid motion reconstruction method. Nowthe non-rigid motion reconstruction algorithm is the factorization method andparameter optimization method two categories. Factorization method is tosimultaneously factorize the tracking matrix into motion and structure matrices bysingular value decomposition (SVD) with a rank constraint. As the motion estimationand structure reconstruction is a non-linear inverse problem it is difficult to find thenon-rigid constraints. Parameter optimization method is bound to rely on closed-formsolution method. The cumulative of Initial value and the objective function noise it iseasy to fall into local minimum. This paper presents a manifold-based group ofnon-rigid motion models created in the non-rigid motion constraints and find animproved non-linear method selecting objective function and the initial value. Thenuse the classical nonlinear optimization methods obtain non-rigid three-dimensionalstructure of the matrix and the parameter matrix for three-dimensional reconstruction.The experimental show that the error of this solution is smaller significantly.(3) Use simulation and experimental validation to test the algorithm. Find the objectfeature points then get the two-dimensional coordinates of feature points from imagesequence. Use the non-rigid motion reconstruction algorithm analysis of results andperformance evaluation algorithm...
Keywords/Search Tags:Non-rigid, Image sequences, Manifold groups of motion, Nonlinear optimization, Three-dimension reconstruction
PDF Full Text Request
Related items