Real Time Reconstruction Of 3D Scene Based On Kinect And Related Technology Research

With the continuous development of computer vision technology,three dimensional reconstruction technology has become a new research hotspot and difficulty.The traditional three-dimensional reconstruction technology based on laser scanner and structured light scanner not only the equipment structure is complex,and not easy to operate,but also is expensive and lacks of real-time performance.These limitations restrict the popularization and development of three-dimensional reconstruction technology,and a new Device that enables low-cost,easy-to-use,high-precision 3D modeling is urgently needed.In response to these problems,this thesis uses the consumer Kinect depth sensor to collect real-time indoor scene data and generate the corresponding three-dimensional model.The specific research contents of this thesis include:(1)To solve the problem of mismatched point pairs in the process of point cloud registration,this thesis proposes a double threshold elimination algorithm.According to the invariance of point cloud geometric features in the process of rigid body transformation,two nearest neighbors of each point are calculated depending on the k nearest neighbors to form a three-points set on the basis of initial matching point set.The approximate congruent 3-points pair is selected according to the invariance of distance between points to finish the preliminary screening.Then the geometric feature of the 3-points sets is described by the surface variation of local area,and the final screening of 3-points pairs is given by the analysis of covariance matrix of the 3-point sets.The experimental results show that the method can effectively eliminate the mismatched points,the registration results are more accurate.(2)Based on the Kinect Fusion algorithm,this thesis constructs a real-time 3D reconstruction system of scene.To solve the problem of the roughness of the vertex estimation in the original system,this thesis estimates the vertex vectors by analyzing the distribution of the point cloud by analyzing the covariance of the sample points and the set of four adjacent sampling points.To solving the problem of corresponding points in the process of solving the transformation parameters has the different contribution,this thesis presents the concept of confidence and given different confidence to matching point pairs in order to improve the stability of camera tracking;And in the process of data fusion,the weight coefficient is optimized and the color information is fused so that the three-dimensional model is more realistic.Finally,the MC algorithm is used to generate a visual three-dimensional color model.