| The light field three-dimensional(3D)imaging system enables a high-quality,high-resolution color dynamic 3D display.It can also achieve complex texture and light shadow.Therefore,in the current situation,light field 3D imaging technology has been studied more and more extensively.Nowadays,the virtual light field sensing and a real light field sensing technology supporting multi-angle graphics rendering,is becoming more and more mature.The display resolution is getting higher and higher,the modulation rates of various light modulators are gradually increasing,besides,the rapid development of computer graphics and image technology is also gradually increasing.So the 3D imaging of a directional light field reconstructed by geometric optical directional screens and projection techniques is being perfected.In recent years,the accuracy and interaction capability of light field 3D imaging,has gradually improved.Moreover,gesture recognition,face detection,and other computer vision functions are then quickly added in the light field 3D imaging system.Also,with the increasing popularity of Virtual Reality(VR)and Augmented Reality(AR)technologies,the light field 3D imaging technology is now combining with human-computer interaction technology,thus increasing the real sensory experience of the viewer.This can provide the viable solution of human-computer interaction for the production of low-cost,high-quality virtual reality and augmented reality.Nowadays,the main work of the light field 3D imaging is real-time updating from the acquisition end to the display end,and also the dynamic processing of the 3D object through hardware adjustment.This mainly includes the hardware processing algorithms and the parallel operations of graphics processors.At present,the current price of core graphics and image projection components is getting lower and lower,but the processing power has not improved significantly with this price advantage.Although graphics processing chips with more advanced functions already exist,the price of new products is still high.The light field 3D imaging technology mainly includes three parts:light field 3D acquisition,light field 3D reconstruction,and light field 3D display.The aim of this thesis is to improve the efficiency of the three-dimensional imaging,reduce the cost,improve the accuracy of the system,and reduce the error.The thesis mainly focuses on three parts related to the technologies of light field 3D imaging:light field 3D acquisition,light field 3D reconstruction,and light field 3D display.The main innovative research contents are as follows:(1)Aiming at the key technology of 3D registration in the light field 3D acquisition technology,an optimization method of an Iterative Closest Point(ICP)algorithm is proposed.The current ICP algorithm is often accompanied by inaccurate initial values for the rotation matrix and translation vector,which increases the number of iterations and takes too long,and cannot be used in multi-object oriented scenes.In order to solve this problem,this thesis proposes a fast 3D registration method which can accurately obtain initial value and can be applied to multi-object oriented scenes.Firstly,the scattered point cloud data is integrated into a complete set of point cloud data through pass-filtering and local surface fitting normal estimation algorithms.The spatial features include geometric features and texture features.Secondly,point cloud segmentation and 3D centroid calculation are used to obtain the point cloud data and their centroids of the regular objects in the scene,such as cups,desks,etc.Thirdly,singular Value Decomposition(SVD)is used to obtain the rotation matrix of each point cloud model.Finally,the translation vector of each point cloud model is calculated by combining the 3D centroid algorithm and the rotation matrix.The experimental results show that compared with the existing ICP algorithm,the proposed method reduces the number of iterations,improves the working efficiency of the system,and solves the inaccurate initial value problem of 3D point cloud registration in the multi-object oriented scenes.At the same time,the registration efficiency of the 3D point cloud of a single-object oriented scene is increased about by 5%compared with the existing method.(2)Aiming at the 3D object recognition technology of light field,an efficient 3D object recognition method is proposed.Efficient object recognition plays an important role in light field 3D reconstruction;especially combining the application of a 3D image acquisition sensor is a new attempt in the field of 3D imaging of light field.Therefore,in order to achieve efficient 3D object recognition in the 3D reconstruction of the light field,a method of 3D object recognition based on Monte Carlo random sampling is proposed,which realizes efficient 3D object recognition in the 3D reconstruction of the light field.The proposed method mainly includes the steps of normal estimation,uniform key point sampling,Monte Carlo random sampling,feature descriptor extraction of SHOT(Signature of Histograms of Orientations),KD-tree index matching,and 3D Hough transform.The proposed method can be implemented only by a single-threaded CPU(Central Processing Unit).In this way,a large number of GPU(Graphics Processing Unit)can be released in the 3D reconstruction stage,and finally,the efficiency of the light field reconstruction is improved.The experimental results show that compared with the traditional methods,the proposed method improves efficiency by 9.26%on average under the same 84.67%correct recognition rate,and it further improves the performance of 3D object recognition in 3D reconstruction,speeds up the recognition process,and provides an effective implementation scheme for the subsequent 3D display of light field.(3)A robust pose estimation optimization method is proposed for the light field 3D display technology.At present,there are still large amounts of hardware resources and manual adjustment parameters needed to realize the pose estimation and display in the light field 3D display.In order to solve this problem,from the perspective of the algorithm,this thesis proposes a method to propose a method to optimize the pose estimation of a 3D display of light field based on H_?optimal control.The proposed method is divided into three parts:firstly,the normal vectors of rigid objects are collected,and the normal vectors of principal component analysis(PCA)are estimated.Secondly,the H_?of the normal estimation result of the Hardy space is calculated,which is the maximum singular value of the rational function matrix of the right half-plane of the complex plane.Finally,the pose estimation and transformation are performed using the optimization results of the H_?.The experimental results based on three international standard databases(Kinect,Mian,and Clutter)show that the proposed method can achieve high quality,high efficiency,and low cost for the robust pose estimation and display of the 3D rigid object in light field 3D display.Also,it is of high accuracy and less time-consuming.In a word,this paper makes an in-depth study on the efficiency improvement,performance improvement,cost reduction,accuracy improvement,and optimal design of light field 3D imaging technology.Compared with the existing methods,the novel algorithms can improve the accuracy and speed of 3D registration,the recognition efficiency of 3D objects,and the accuracy and robustness of 3D rigid object pose estimation.These can be used to further study the combination of the prior knowledge of the scene acquisition and the reverse derivation reconstruction decoding,high-resolution and high-dimensional spatial information image acquisition,and 3D object vertex position coordinates and shader loading methods provide research results and reference. |
PDF Full Text Request