Research And Development Of Personalized Human Body Modeling Using Kinect

3D modeling of personalized human body is a method for building3D geometryshapes to approximate the users in virtual scenes. Human body modeling is a hot topicwhich consists of techniques on Computer Graphics, Computer Vision and Engineeringof Human beings and promoted by these techniques compactly. The pertinent researchin this area would be exploited to build avatar for users with scanning data, and alsolever the interaction experiences in virtual environment to a higher stage, finally drivethe rapid development of human modeling in animation, filming, virtual reality and soon.After evaluating the applicational prospect of modeling technique for personalizedhuman body, as well as summarizing current modeling methods adequately, this paperbrings forward a modeling technique which is relied on Kinect, a novel depth cameraequipment, to capture shape data, and also makes use of prior knowledge to create thehuman mesh model which approximates to real scanning data. The major work consistsof following three steps:(1)As the inherent drawbacks of Kinect, the captured data is low-resolution andnoisy, so the first step is preprocessing of data to smooth noise;(2)Being influenced by incompleteness and noise of scanning data, SCAPE modeland semantic analysis method are utilized to learn the prior knowledge in human models.Then some correlated constraints are established to fulfill the modeling of personalizedhuman body.(3)After exploiting PCA to extract some decisive parameters for human shapeapproximation, average mesh model is created and fitted to the incomplete noisy modelcaptured by Kinect. Subsequently, personalized mesh model is reconstructed.This paper takes advantage of the priorities in low expense and interactionalconveniences of novel depth camera, such as Kinect. Meanwhile, it combines theexcellence in model-driven, template-based and scanning-data modeling techniques, andsolves the problems in modeling of personalized human body with fusion of abovemethods. The experiments show that we can create correct and high quality meshmodels for human being with single view scanning data. The reconstructed models arewith high fidelity in human parameters, such as2%error in height.