針對目前三維重建技術不方便應用於人們日常生活，而且計算量大的問題，提出一種基於深度和彩色圖像的人體三維點雲配准方法，首先通過Kinect設備獲得不同視角下人體RGB圖和深度圖，採用SIFT和RANSAC演算法對RGB圖像進行特徵點對提取並映射到三維坐標系，完成粗配准，採用改進的ICP演算法進行精配准，最後對點雲進行背景和模型邊緣雜訊剔除，實驗結果證實，該演算法在人體正面與側面曲率變化大，重疊區域小的情況下能夠比較準確地完成人體點雲的配准。
The existing 3D reconstruction techniques rarely can be easily used in daily life, and the traditional registration algorithm has the drawback of massive calculation. In this paper, a registration algorithm of 3D body point cloud based on depth and RGB images is presented. First, the Kinect is used to obtain the depth and RGB information of human body from different perspectives. Then the corresponding pairs of 2D feature points are extracted from RGB images by SIFT and RANSAC algorithm in the coarse registration. Afterwards, the improved ICP algorithm is available for the fine registration. Finally, the background information and the noise points of the model edges are eliminated from the image. Experimental results show that the body point clouds can be accurately and efficiently registrated by the proposed algorithm with economical instrument.
