The Research Of Shape Alignment Problem In 2D And 3D Space

Shape alignment technology is one of the popular research fields in computer vision and computer graphics.The main problem need to be solved is to align two or more objects in space fastly and accurately,in order to the shape alignment technology can be better applied in the restoration,graphic retrieval,interactive games,3D scene reconstruction,and other areas of the application.the problem faced in the process of shape alignment are: first,the input of model images are often inaccurate,as in the contour of the two-dimensional space is digital image and noise,therefore,we need to preprocess the input image and select the right shape attribute to reduce the effects of noise;On the other hand because of the large number of points of the model in the calculation of shape alignment the efficiency of solving similarity measure is very low.so a suitable algorithm need to proposed to accelerate shape alignment process.This paper will construct a mathematical model with constraint to solving the problems of the shape alignment in space.We study the shape alignment respectively based on 2D and 3D space,including the preprocessing of the shape alignment,the similarity calculation of shape alignment and application based on shape alignment.In this paper,the main research results are:(1)In the shape alignment of two-dimensional,the key of the problem is that using contour features or contour point attribute to calculate the best public section.Considering the existing shape when looking for a best alignment section alignment method is sensitive to noise and with low calculation efficiency shortcomings,we propose a fast calculation method about 2D contour alignment based on local EMD(Earth Movers Distance)measurement in this paper.The method adopts the equidistant take as alignment points in the contour line,and then calculate the starting point and end point for contour alignment with the help of the EMD algorithm,here the properties of the length,color and curvature of point set are calculated at the same time;In a number of 2D contour alignment experiment,this paper presents a greedy algorithm is used to improve the efficiency of alignment.Map matching and the restoration of experiments show that this algorithm is fast,is not sensitive to noise,and can be applied to 2D shape alignment well.(2)In 3D shape alignment,this paper studies the problem of automatic alignment based on the multiple kinects.In this paper,the specific process of three-dimensional skeleton data registration based on kinects is: firstly,we get the human body skeleton data with kinect,here to capture 3D bone data contains the 25 points of human body skeleton;Then use the L-BFGS algorithm to calculate transformation matrix from the random skeleton data to standard data.In order to reduce the number of parameters in this paper,we use the quqter to represent the transformation matrix;At the end of the paper further optimization will be done based on the dynamic sequence of transformation matrix to obtain stable accurate transformation matrix.In the alignment of multiple skeleton data we get the standard skeleton at the same time.Compared with traditional kinect alignment methods,our method does not need to use objects to complete alignment,and improve speed and accuracy of the kinect alignment.