Sequence Reconstruction Of Flexible Deformable Objects Based On RGBD Images

Object reconstruction technology is widely used in animation,garment CAD,medical imaging and other fields.With the popularity of RGBD cameras,RGBD-based 3D reconstruction is becoming the mainstream of current research.However,the current 3D reconstruction technology isn't completely suitable for the flexible deformable objects.Due to the self-occlusion and other factors,the results of single 3D reconstruction are incomplete.And the easy-to-deform properties of flexible objects can not be well reflected.In this paper,the dynamic reconstruction of flexible objects in deformation is studied to capture the deformation properties of flexible objects.A sequence reconstruction scheme of flexible deformable objects based on self-improve template is proposed.A self-improve template is a 3D template that refines geometric information form deformed flexible objects continuously.Sequence reconstruction refers to reconstructing 3D geometric models at multiple times,and realizing the uniformity of grid structure between different 3D geometric models.On the one hand,these models can express more complete information of object,on the other hand,they can also express dynamic deformation information of flexible objects,showing the geometry and physical properties of the objects better.The main research contents can be noted as following:A 3D reconstruction scheme based on multi-camera is implemented.The object model is reconstructed at multiple times with multiple RGB-D(color-depth)cameras.A multi-angle data acquisition system with a circular array is built to collect the surface information of flexible object.3D reconstruction includes data preprocessing,point cloud fusion,mesh generation and texture generation.There are some problems in current 3D reconstruction scheme:high noise in depth image,confusion of foreground and background,unclearness of texture.The following improvements have been made.The average image is used for image filtering,the details can be better preserved while removing noise.The background is segmented using region growing algorithm,the foreground object can be extracted accurately.The texture is reconstructed using ABF++ algorithm,the color information is extracted from the multi-angle color image to generate a continuous and clear texture.A sequence reconstruction scheme based on self-improve template is proposed.The information in a single model from 3D reconstruction is incomplete,but more models of flexible object in different deformation are reconstructed,the information contained is more complete.Therefore,the self-improve template is proposed to integrate information in multiple models.Initializing the template with one model,then merging geometry and color information captured in deformation into the template,the shape and color information of the template is improved.Then a sequence of models with complete information and consistent structure is generated by fitting them with the template,realizing sequence reconstruction.Taking a typical flexible object-clothing as an example,the clothing is deformed with a robot mannequin which can imitate different human body shapes,illustrating the sequence reconstruction.And the application of sequence reconstruction is explored,a simple virtual fitting system is designed.By establishing correspondence between human body measurements and clothing models,the system can generate the deformation effect of the clothing on human body,after the user enters a set of human body measurements.