Font Size: a A A

Dynamic Human Body Modeling Using A Single RGB Camera

Posted on:2018-09-24Degree:MasterType:Thesis
Country:ChinaCandidate:H Y ZhuFull Text:PDF
GTID:2348330512990766Subject:Electronic and communication engineering
Abstract/Summary:PDF Full Text Request
Although 3D human body modeling has a wide application in virtual reality,3D games and many other fields,it is a constant for researchers to explore how to con-veniently obtain high quality huaman bodels with low cost,due to the complexity of human motion.The state of art approach to obtaining high-quality human models is through lasers or depth cameras,which are expensive and clumsy.Thus,some researchers obtained 3D human model through multi-view reconstruction.However,it is not only to cali-brate the position of the camera but also to synchronize all the cameras.In addition,more cameras contribute to high costs,which limits to promoting the multi-view recon-struction system.With the improvements of mobile phones and other mobile devices'imaging quality,how to acquire high-quality human model by single RGB camera has been a hot topic between researchers.Due to the complexity of human motion and occlusion,it is very challenging for us to reconstructe human models using a single RGB camera.At present,most of the methods based on single-camera reconstruction need to add some constraints artificially,and thus can not reconstructe human model automatically.During the human motion,points of human model in a small region have simi-lar motion.And with continuous movement of the human body,it is more possible to capture missing informations through the camera.We design a kinematic classifi?er to vivide the moving body into many "blocks" based on the motion information,and the deformation of the "block" surface in a short period of time was negligible,thus converting the non-rigid reconstruction problem into a rigid reconstruction prob-lem.We fuse the human body part model obtained at different times to the SCAPE moaei.Finally,a anvabie numan moaei will be obtainea after aecomposing the "rigid component" and "non-rigid component" of SCAPE instances.The "rigid component"of the model describes the body's body parameters,while the "non-rigid component"describes the details of the human body at different times.Experiments show that our single-camera dynamic human body reconstruction system not only has high robustness,but also the first system that can automatically reconstruct the complete human three-dimensional model without artificial constraints.
Keywords/Search Tags:SCAPE, non-rigid reconstruction, single RGB camera, motion classifica-tion, structure from motion
PDF Full Text Request
Related items