3D Models Reconstruction And Optimization

3D models reconstruction is one of the core contents in computer graphics research.It is widely used in industrial manufacturing,product design,film and television animation production,virtual reality and many other fields.With the increasing complexity of research objects,traditional 3D modeling methods have encountered enormous challenges in terms of computational efficiency and interactive naturalness.3D models reconstruction of real-world objects is based on visual information such as images,depth,and video,which is greatly improving the automation level of the modeling process.Meanwhile,in order to make the model data meet with the relevant requirements in research and manufacturing fields,it is urgent,but still remains to be a challenge to simplify the users’ operation,reduce the difficulty of modeling,improve the modeling accuracy and obtain models’ semantic information.Based on the above related research,my dissertation mainly focuses on imagebased interactive modeling and obtaining the complete semantic information of construction models.Our study effectively reduces the modeling threshold,improves the modeling efficiency,and automatically extracts,analyzes and optimizes the complete semantic information of the model,thus obtaining a “readily usable” 3D model.This thesis focuses on how to obtain and optimize model semantic information.The semantic information is consisted of two levels: High-level information,which contains geometric data and motion parameters of the model components;Low-level information,including the mesh topology of the 3D model.Accurate model geometry data is the basic information for building large-scale virtual digital scenes.Complete model motion parameters enable high-quality model motion simulation and evolution.High-quality model mesh topology accelerates algorithmic efficiency in model simulation and transmission.In general,the research content in this paper is to obtain and optimize the complete 3D model information,so that the reconstructed 3D models can be directly applied to a wide range of industrial and research fields such as 3D printing,virtual reality,and motion simulation.The main contributions of this dissertation are as follows:(1)An efficient and accurate “simple” interactive modeling method based on multi-view images is proposed.Aiming at solving the problems of complex 3D modeling process,high requirements for user’s professional knowledge and poor natural interaction,this paper proposes a modeling interaction method for directly drawing out model sketches through simple “three strokes” on multi-view photos,which greatly improves the modeling.The interactive nature of modeling reduces the threshold for modeling.On this basis,to solve problems of occlusion between the model parts and poor modeling accuracy in the image,the point cloud data obtained from the multi-view image of the model can be used to assist the users to draw the model.Based on the geometric constraint relationship between the data-driven detection model components,the geometric parameter information of the model components can be optimized,and finally an accurate 3D model that exactly matches with the real photo can be obtained.Based on these research results,we developed an interactive 3D modeling system based on multi-view images.For common users,within 30 minutes,a complex 3D mechanical model with industrial precision such as “Poston-engine” and “Robot-arm” can be reconstructed.(2)A method for automatic motion extraction and optimization of 3D model based on video sequence is proposed.The professional threshold for the artificial construction of 3D motion models is quite high and also it is a heavy workload for artificially construction of 3D motion models.Meantime,as for the motion model parameters acquired from video automatically,the motion accuracy is quite low and cannot deal with complex motion chains.Aiming at these problems,this paper proposes a motion joint type search algorithm based on mechanical motion prior knowledge and a model motion matching optimization algorithm based on video sequence.In this paper,based on the prior knowledge of mechanical model joints,a joint set of different joint types has been constructed and a transfer chain for model motion has been established.The energy function has been established to match with the 3D model silhouette and the image boundary.Through the simulated annealing optimization of energy function,the joint type of the model and the motion parameters of the component are optimized in order to obtain the model motion information that completely matches with the real motion in the video sequence.This method is a good solution to the difficulty of formal description and optimization of complex model joint motion.According to the experimental results,the reconstruction model obtained by 3D printing has the same motion result with the video.(3)A fast disassembly and packing method for 3D models based on component motion optimization is proposed.For the 3D model,when the 3D printing is performed by fused deposition,the printing time is long and the supporting material is expensive.In this paper,by optimizing the motion parameters of the model components,we greedily split the joints at where the space cost is highest,and then calculate the component parameters which can minimize the bounding box volume of the model component group,so that the model can be arranged more compactly and reach the minimum bounding box volume in 3D printing.Since model disassembly and packing are both NP-hard problems in dynamic optimization,this paper adopts hierarchical splitting strategy and heuristic packing strategy to effectively balance the optimal number of disassembling and packing space utilization.Meantime,this method can improve the packing efficiency compared with the traditional packing algorithm.(4)A parallel and robust semi-regular triangle remeshing method is proposed.To solve the problem of slow rendering speed and low grid compression rate,the semiregular hierarchical structure(Levels of Detail,LOD)is usually used to express complex models with multiple resolutions.The current semi-regular method algorithm is inefficient and cannot meet with real-time requirements.In this paper,the layered simplification and the establishment of parallel independent sets effectively solve the problem that the mesh simplification algorithm can’t be paralleled,so that the algorithm can achieve real-time efficiency in parallel.The simplified rough model mesh can be subdivided into precise semi-regular meshes by parameterized smoothing and highprecision sampling during mesh simplification.