Research On Physics-based Data-driven Geometric Modeling

The rapid development of informatization in current society makes human easily and quickly acquire the data that they need in many different ways.However,except for data acquisition,each of us also generates data all the time,which makes the topics about big data very hot.But in computer graphics field,the era of big data still seems to be far from blooming,which is mainly due to the cost of data modeling.Therefore,the related data-driven methods have not attracted much attention.But nowadays,as scanning device and wearable device are becoming more and more popular,we believe in that the big data era in computer graphics is coming soon and data driven method will become more and more important.The popularity of scanning device brings huge data for geometric research.However,a large mount of the data is raw data and cannot be directly dealt with via most geometric algorithms.Thus,due to the short of motion data,in this paper we have researched on motion data generation.After that we also have done research on how to make use of the motion data.In specific,we first propose an automatic method that interpolates two gestures into a series of motion data with respect to time.Based on such motion data,clothes have been simulated as training data.After the analysis of the training data,when given a new gesture,suitable clothes can be efficiently generated.In addition,we also exploit physical analysis on models for 3D print,which can help print objects with different properties using only a single material.The contributions of the paper include:1.Shape interpolation has been widely used in many fields.It mainly produces seamless transition data that deforms one object into another.From another view,shape interpolation could be regarded as a way to generate new data.In order to produce satisfactory in-betweens,the most important is to establish accurate correspondences between source and target objects.It is a hard problem in computer graphics.In this thesis,we introduce an automatic 3D shape interpolation method based on skeletons to produce smooth in-betweens.Based on as-rigid-as-possible strategy,our method can provide plausible morphing results.2.Clothes simulation is also an important area in computer graphics and it has great demand in many fields.In this paper,we present an automatic method to realistically animate garments on human poses in reduced space.Our framework can be mainly divided into offline phase and online phase.During the offline phase,based on linear blend skinning,rigid bones and flex bones are estimated for human bodies and garments respectively.Based on their correspondences,we can learn their relations.In the online phase,new human poses are traded as input to estimate rigid bone transformations.Then both rigid bones and flex bones are used to drive garments to fit the new poses.Experiments manifest that our method is very fast and results are realistic.3.Most frequently used 3D printer can print objects using only a single material,which greatly limits its application.In this paper,as for such situation,we adopt microstructures with different properties to construct the target object.Then we print such target object using only a single material to achieve the goal that the target object can have desired behaviors.Besides,we also build a platform to scan and track real objects,and then use related algorithm to estimate objects' base material properties.