Font Size: a A A

Non-rigid Deformation Space Building With RGBD Video

Posted on:2019-03-14Degree:MasterType:Thesis
Country:ChinaCandidate:Z H WangFull Text:PDF
GTID:2428330548979817Subject:Computer Science and Technology
Abstract/Summary:PDF Full Text Request
3d deformation techniques have been the subject of many studies,these technologies are also widely used in the animation and game.With the popularity of consumer-level depth cameras,more and more research and application based on depth camera has appeared.We propose a deformation subspace technique based on RGBD video.With the input of the RGBD video collected by kinect,we implement the pipeline of 3d reconstruction,object pose es-timation,deformation capture and deformation subspace construction.The modification method based on subspace ensures that the modified models conform to the deformation properties of the object.Our pipeline can be divided into four steps:First of all,we scan the object and reconstruct the 3d model.Then we will take the RGBD image as input,according to the operator's hand posi-tion to segment the part belonging to object from the input image,and estimate the initial pose of the object.This step can be regarded as the pretreatment of deformation capture.Then we will use the depth video as the input,describe the deformation of the model with a deformation graph based on the dual quaternion,capture the deformation of the object by optimizing the deformation graph parameters,and pick representative deformation state as a deformation keyframes.Finally,we ex-tract the feature vectors based on deformation gradient from these deformation keyframes,and use these feature vectors as the basis vectors to span the nonlinear feature space.The deformed model corresponding to the feature vectors in the feature space constitutes the deformation subspace of the model.In addition,we also give a shape modification method based on the subspace to ensure that the modified model falls within the subspace.Because 3d models and deformation keyframes are all obtained by capturing the surface in-formation of the object with an RGBD camera,not obtained through manual editing,it is to ensure that the static 3d model and the deformation keyframes are consistent with the properties of the object.Therefore,the resulting deformation subspace is reasonable.
Keywords/Search Tags:RGBD-video, dual quaternion, deformation capturing, deformation subspace
PDF Full Text Request
Related items