Vision-based Extraction Of Human Body’s Geometrical Information And Model Reconstruction

With the development of Internet technology and 020 e-commerce mode,more and more consumers are no longer satisfied with buying clothes online without fitting.Then,the concept of virtual fitting and aided tailoring emerged.Virtual fitting technology can achieve the effect of fitting without taking off,and the aided tailoring technology can complete individualized tailoring.The realization of these two technologies generally requires the geometrical information of user’s body as a basis.The information include the feature points’ positions which reflects the information of body’s posture and body sizes which reflects the information of body’s morphology.By extracting the information,body can be parameterized for 3D reconstruction of the human body.Depending on the difference of data acquisition device,we’ll study two algorithms of vision-based extraction of human body’s geometrical information respectively.The data source of RGB-image-based extraction of human body’s geometrical information is the body’s front and side RGB images.To extract a precise result of body from the image,the algorithm defines that the background should be concolorous and the human body should be non-occluded.For the front image,the algorithm firstly finds the approximate location of the human body with the technology of face detection quickly,obtains accurate body contour with the technology of edge detection,designs and uses the efficient method of orthogonal list storage and retrieval to find the key points of the contour,and finally gets the feature points’ positions and front body sizes.For the side image,the algorithm uses the same way to store contour,and get the side body sizes according to the side morphological characteristics of the human body.Finally,the result of the algorithm is applied to the human body’s model reconstruction,and experiments show that the algorithm has better instantaneity and accuracy.Although capturing RGB image is inexpensive and convenient,RGB-image-based algorithm generally adapts to limited postures only and can’t remove complex background well.So it always can’t well serve the applications that need human models.To solve those problems,we present an algorithm of RGBD-image-based extraction of human body’s geometrical information which only needs one Kinect.It detects the feature points of human in depth video frames combining with error correction firstly.Then,it measures or estimates body sizes,and improves them progressively while processing the video stream.Finally,it achieves the segmented parametric modeling with above parameters and generates the human model in a whole.The experiments demonstrate that our method can reconstruct the 3D human with incomplete sizes.The semantic information carried in the model helps to achieve the convenient partial modification,good efficiency and interactive effect in the virtual dressing application.