【作者】 于子宸； 刘全利； 王伟；

【Author】 YU Zi-chen;LIU Quan-li;WANG Wei;School of Control Science and Engineering, Dalian University of Technology;

【机构】 大连理工大学控制科学与工程学院；

【摘要】 场景流估计是当前科学界与工业界均给予关注的热门领域,在三维场景理解、场景动态重建等领域具有重要的作用。本文在考虑相机移动的情况下,提出了一种改进的基于RGB-D图像序列的场景流估计算法,该算法能够对相机运动和场景流进行联合估计。算法的核心是对场景进行双重分割,首先利用本文提出的基于K-Means的场景分割优化算法将场景划分为几何簇,将每一簇视为刚体,极大的简化了场景流估计问题;然后将场景分割为背景区域和运动区域,所有分割为背景的簇被用于相机运动的细化,而每一个运动簇单独计算其对应的刚体运动,进而得到其对应的场景流。将本文提出的算法与几种最先进的场景流估计算法进行比较,结果表明,该方法能够在保持较高实时性的同时提高场景流估计精度。更多还原

【Abstract】 Scene flow estimation is a hot area in both scientific and industrial fields, which plays an important role in 3 D scene understanding and scene dynamic reconstruction. Considering the situation of camera movement, this paper proposes an improved scene flow estimation algorithm based on RGB-D image sequence, which can jointly estimate the camera motion and scene flow. The key of the algorithm is the two-fold segmentation of the scene. First, the scene segmentation algorithm based on K-Means is used to divide the scene into geometric clusters and treat each cluster as a rigid body, which greatly simplifies the estimation of scene flow. Then, the scene is divided into the background region and the moving region. All the clusters divided into the background are used to refine the camera motion, and each moving cluster calculates independently its rigid motion to obtain the corresponding scene flow. Comparing the proposed algorithm with several state-of-the-art methods in scene flow estimation, the results show that the method can improve the accuracy of scene flow estimation while maintaining high real-time performance.更多还原