Research And Application Of 3D Reconstruction Technique Based On Depth

The three-dimensional reconstruction uses a plurality of two-dimensional images or depth cameras and other apparatus to determine the spatial position information and the three-dimensional geometric shape of the object.Because of its practicability,it can accurately restore the characteristics of real scenes or objects,which is widely used in medical,cultural relics,national defense and other fields.However,at present,the three-dimensional reconstruction still has the disadvantages of tedious and inaccurate acquisition of three-dimensional information of objects,and it cannot meet the large-scale demand for three-dimensional information and three-dimensional models of contemporary people yet.Therefore,it has become a hot research direction in computer vision technology that how to obtain the three-dimensional information of high precision objects or scenes,and to efficiently reconstruct the three-dimensional model of the real effect of the object or scene.With the rapid development of computer vision technology and continuous improvement of hardware devices,a large number of depth sensing devices appear in front of the public,such as Microsoft's Kinect and Asus Xtion PRO,which has opened up a new way for the research of 3D reconstruction.In this paper,Kinect depth sensor is used to study the 3D reconstruction technology of depth images and applied into automation equipment modeling.Several key technologies in 3D reconstruction are studied in this paper.Firstly,get the three dimensional information of object.The Kinect is used to scan the reconstructed object and obtain its depth information,then transform the two-dimensional depth image into three dimensional point cloud data,so as to obtain the 3D point cloud information of the reconstructed object surface.Secondly,improve chaotic glowworm swarm optimization particle swarm optimization.The earlier stage of this algorithm is aimed at solving the problem with uneven distribution of initial individuals of the artificial firefly algorithm,and a Logistic map introduced to initialize the population.And the adaptive step length method of a dynamic grouping strategy is used to coordinate the search step of the algorithm.The improving algorithm is combined with the particle swarm algorithm in the later stage.The initial population of the particle swarm algorithm is generated by the result of each iteration optimization.In order to solve the problem of the accuracy of solution of glowworm swarm optimization is low,the initial population is then searched for the global optimal solution by using the particle swarm optimization algorithm with high convergence speed and high accuracy.The improved algorithm is tested by function.The results show that the convergence rate of improved algorithm is superior to that of standard particle swarm optimization and artificial firefly algorithm.Thirdly,optimized iterative closest point algorithm(ICP).First of all,for the problem of the ICP algorithm to search the corresponding point for slow speed,the initial corresponding point pair is determined by kd-tree to improve the search efficiency.Then,owing to the noise point is mismatched and the algorithm is easy to fall into the local optimal problem,the distance weight and the feature weight restriction strategy are used to remove the noise points and mismatch points,and the registration precision is improved.Finally,combined with method of the initial registration and fine registration of the point cloud are carried out by ACGPSO-ICP algorithm,which solves the local optimal problem of the ICP algorithm that the initial position of the point cloud is different,which ensures the accuracy and robustness of the algorithm.Practical application shows that this method improves the registration accuracy and efficiency of 3D point cloud data.Finally,the 3D reconstruction of the commercial installation.On the basis of the previous research,the depth image of the Kinect depth sensor is obtained and transformed into a point cloud.By the initial registration and fine registration of the three dimensional point cloud,the three-dimensional point cloud is converted to the same coordinate system.Then the point cloud is triangulated by Greedy projection triangulation algorithm to get the complete three-dimensional model,and finally the 3D reconstruction is achieved.