Research On Reconstruction, Segmentation And Dimension Extraction Of3D Human Model

With the rapid development of digital information and application, virtual reality of the3D digital human body becomes more and more widely in production and life.3D point cloud data is collected by the non-contact laser scanning the body surface. The3D body model can be use for medical digital people, garment design and production,3D movies and games and other fields of intelligent direction research in the computer processing. The3D body model is reconstructed based on3D point cloud data in this paper. The segmentation algorithm of model and the extraction technology of key size are discussed.The3D human model reconstruction needs to make coordinate fusion, filter some data points, remove the noise points, and normalize. The original data points are obtained by the non-contact laser scanning the surface of the body directly.After completing the model reconstruction, the whole model is divided into6parts, including head, limbs and torso. For the segmentation of head from torso, the method of maximum distance combined with approximate edge of convex hull is proposed. Using the approximate convex hull edge to locate the feature points can reduce the amount of data calculation. In addition, an oblique angle of cut section is introduced, and the method improves accuracy of head segmentation. For the segmentation of lower limbs from torso, the method of iteration curve fitting is proposed based on hip contour. Use the iteration curve fitting method to segment lower limbs rather than the fork point method. In this way, it can remove noise points in the contour effectively and improve accuracy of lower limbs segmentation. For the segmentation of upper limbs from torso, the algorithm is proposed based on concave features points in horizontal slices. There are four concave feature points in the every horizontal slice, which is between the bifurcation point of arm and torso and the endpoint of shoulder. According the four concave feature points can divide upper limbs form torso in every slice, and avoid the armpit point located inaccurately. The ultimate goal of the human body is to extract sizes form the body surface after segmenting the model. There are36dimensions to show the human body morphology according to the GB/T16160-2008. In addition, some basic clothing sizes are extracted based on clothing mannequins with triangle of grid pattern.Based on above researches, this thesis designs and implements a human dimension automatic measurement application system. The system can rebuilt model with3D point cloud data, and the model can rotate arbitrary angle in the horizontal direction or vertical direction. Every one of six parts can be choose to observe in this system real-timely, and the values of36key dimensions can be extracted accurately.