基于RealSense人体多视角三维重建方法的研究

Research on multi-view 3D reconstruction method of human body based on RealSense

深度相机扫描人体的数据往往存在噪声以及相邻两帧配准存在误差的问题。提出了一种基于RealSense的人体三维重建方法。采用RealSense传感器获取不同视角下的多片点云,经过预处理后得到主体点云;结合快速点特征直方图(FPFH)与随机抽样一致性(RANSAC)对主体点云粗配准;改进迭代最近点算法对主体点云精配准,最终得到精准的人体三维模型。实验结果表明,该方法可以快速重建具有真实感、细粒度和尺寸精准的三维人体模型。

The data of the human body scanned by the depth camera often has the problems of noise and registration errors of two adjacent frames. We propose a three-dimensional human body reconstruction method based on RealSense. RealSense sensor is used to obtain multiple point clouds from different perspectives, and the subject point cloud is obtained after preprocessing;combined with fast point feature histogram(FPFH) and random sampling consistency(RANSAC) to coarsely register the subject point cloud;initialize with weights The method improves the iterative nearest point algorithm to precisely register the subject point cloud, and finally obtains an accurate three-dimensional model of the human body. Experimental results show that this method can quickly reconstruct a realistic, fine-grained, and accurate three-dimensional human body model.