3D Reconstruction of Human Body in Virtual Fitting Room Based on Kinect

Abstract

: With the progress of the times, more and more scientific and technological elements have been integrated into people’s daily life, which is manifested in fitting. Virtual fitting technology provides people with a more convenient and interactive fitting mode. The launch of Microsoft Kinect solves the problem of human body spatial information acquisition and facilitates the development of virtual fitting systems. This paper uses modeling software to build a human body 3D clothing model, and focuses on the human body 3D clothing modeling. This paper binds the three-dimensional clothing model with human bones to the user’s three-dimensional information collected through the Kinect camera to achieve the fusion of virtual and virtual clothing. This paper simulates the physical characteristics of clothing fabrics to improve the realism of virtual clothing degree. The iterative nearest point algorithm is improved. First, the voxel grid is down-sampled for the two point clouds, and then the scale-invariant feature points of the source point cloud are found and saved as a point cloud. The saved point cloud is registered with the target point cloud sampled from the voxel grid. In this paper, the human body point cloud data is collected through Kinect, and the point cloud segmentation, point cloud registration and point cloud reconstruction are studied separately, which makes the Kinect-based 3D human body modeling method more efficient and accurate. This paper proposes a method of iteratively deforming the standard model using the mesh deformation migration algorithm. The method is to establish a mapping relationship between models by given a set of corresponding point pairs between the source grid and the target grid, and realize the constrained deformation from the source grid to the target grid. Experiments show that the algorithm proposed in this paper uses a cheap depth camera to scan the human body. The algorithm preprocessing time is only about 1 second, and the average optimization time is about 3.6 seconds. It can overcome the shortcomings of low depth camera data accuracy, and the reconstruction time is short and the result is high accuracy.