粗匹配和局部尺度压缩搜索下的快速ICP-SLAM

发布时间：2018-03-14 03:12

本文选题：ICP-SLAM　切入点：粗匹配　出处：《智能系统学报》2017年03期 　论文类型：期刊论文

【摘要】：ICP-SLAM在自主机器人和无人驾驶领域得到了极大的关注,但传统ICP-SLAM缺少当前帧和全局地图的相对位置关系,因此本文ICP算法必须经过大量的迭代之后才能达到收敛条件,这导致传统ICP-SLAM实时性很差。并且在每一次的迭代过程中,必须通过全局搜索才能完成匹配点搜索,这进一步降低了传统ICP-SLAM的实时性。为此,提出了一种快速ICP-SLAM方案。首先,通过MEMS磁力计和全局地标计算出初始位姿矩阵,通过该初始位姿矩阵实现当前帧和全局地图之间粗匹配,进而减少达到收敛条件的迭代次数。其次,在每次迭代过程中,将采用局部尺度压缩搜索完成匹配点搜索,从而减小ICP-SLAM的计算开销,提高ICP-SLAM实时性;同时,每次迭代完成之后,还将通过动态阈值缩小搜索范围,达到加快匹配点搜索的速度,进而提高ICP-SLAM实时性。实验结果表明,和传统ICP-SLAM相比,在理想室内静止场景下,快速ICP-SLAM的迭代次数最高减小了92.34%,ICP算法运行时间最高降低了98.86%。除此之外,ICP-SLAM的整体负载也被保持在可控范围内,ICP-SLAM的整体性能得到很大的提升。
[Abstract]:ICP-SLAM has attracted much attention in autonomous robot and driverless field, but the traditional ICP-SLAM lacks the relative position relation between the current frame and the global map. Therefore, the ICP algorithm in this paper has to go through a lot of iterations before it can reach the convergence condition. This leads to the poor real-time performance of traditional ICP-SLAM, and in each iteration process, the matching point search must be completed by global search, which further reduces the real-time performance of traditional ICP-SLAM. Therefore, a fast ICP-SLAM scheme is proposed. The initial pose matrix is calculated by MEMS magnetometer and global landmark. The rough matching between the current frame and the global map is realized by the initial pose matrix, and the number of iterations reaching convergence conditions is reduced. Secondly, in each iteration process, The local scale compression search will be used to complete the matching point search, which will reduce the computational overhead of ICP-SLAM and improve the real-time performance of ICP-SLAM. At the same time, after each iteration is completed, the search scope will be narrowed through dynamic threshold, and the speed of matching point search will be accelerated. The experimental results show that compared with the traditional ICP-SLAM, in the ideal indoor static scene, The maximum iteration time of fast ICP-SLAM is reduced by 92.34 and the maximum running time of ICP-SLAM is reduced by 98.86. in addition, the overall load of ICP-SLAM is kept in a controllable range and the overall performance of ICP-SLAM is greatly improved.
【作者单位】：上海大学通信与信息工程学院;上海大学微电子研究与开发中心;
【基金】：国家“863”计划基金项目(2013AA03A1121,2013AA03A1122) 上海市教委重点学科资助项目(J50104)
【分类号】：TP242

【相似文献】