计算机应用 ›› 2017, Vol. 37 ›› Issue (5): 1445-1450.DOI: 10.11772/j.issn.1001-9081.2017.05.1445

• 计算机视觉与虚拟现实 • 上一篇    下一篇

基于完整可见性模型的改进鲁棒OctoMap

刘晙1, 袁培燕2, 李永锋3   

  1. 1. 河南工学院 计算机科学与技术系, 河南 新乡 453002;
    2. 河南师范大学 计算机与信息工程学院, 河南 新乡 453007;
    3. 96367部队, 青海 德令哈 817099
  • 收稿日期:2016-10-13 修回日期:2017-01-10 出版日期:2017-05-10 发布日期:2017-05-16
  • 通讯作者: 刘晙
  • 作者简介:刘晙(1981-),男,湖北天门人,讲师,硕士,主要研究方向:移动机器人视觉定位与地图创建;袁培燕(1978-),男,河南邓州人,副教授,博士,CCF会员,主要研究方向:移动机会网络;李永锋(1989-),男,河南焦作人,硕士,主要研究方向:移动机器人视觉定位与地图创建。
  • 基金资助:
    国家自然科学基金资助项目(U1404602);河南省高等学校青年骨干教师计划项目(2011GGJS-198);河南省教育厅科学技术重点研究项目(14A520045)。

Improved robust OctoMap based on full visibility model

LIU Jun1, YUAN Peiyan2, LI Yongfeng3   

  1. 1. Department of Computer Science & Technology, Henan Institute of Technology, Xinxiang Henan 453002, China;
    2. College of Computer & Information Engineering, Henan Normal University, Xinxiang Henan 453007, China;
    3. Troops 96367, Delingha Qinghai 817099, China
  • Received:2016-10-13 Revised:2017-01-10 Online:2017-05-10 Published:2017-05-16
  • Supported by:
    This work is partially suppotted by the National Natural Science Foundation of China (U1404602), the Funded Project of Key Teachers in Colleges and Universities of Henan Province (2011GGJS-198),the Science and Technology key research project of Education Department of Henan Province (14A520045).

摘要: 从移动机器人自主导航对3D地图精度的需求出发,在鲁棒OctoMap的基础上提出一种基于完整可见性模型的改进鲁棒OctoMap并应用于基于Kinect的RGB-D同时定位与地图创建(SLAM)中。首先,通过考虑相机和目标体素的相对位置关系及地图分辨率进行可连通性判断,获得满足可连通性的相邻体素的个数及位置;其次,根据不同的可连通性情况分别建立目标体素的可见性模型,从而构建普适性更强的完整可见性模型,有效克服了鲁棒OctoMap可见性模型的局限性,提高了建图精度;再次,使用基于高斯混合模型的Kinect深度误差模型代替简单深度误差模型,进一步克服传感器测量误差对地图精度的影响,降低了地图的不确定性;最后,结合贝叶斯公式和线性插值算法来更新八叉树中每个节点的实际占用概率,从而构建基于八叉树的立体占用地图。实验结果表明,所提方法有效克服了Kinect传感器深度误差对地图精度的影响,降低了地图的不确定性,其建图精度较鲁棒OctoMap有明显的提高。

关键词: 3D地图, 鲁棒OctoMap, 高斯混合模型, 八叉树, 立体占用地图

Abstract: An improved robust OctoMap based on full visibility model was proposed to meet accuracy needs of 3D map for mobile robot autonomous navigation and it was applied to the RGB-D SLAM (Simultaneous Localization And Mapping) based on Kinect. First of all, the connectivity was judged by considering the the relative positional relationship between the camera and the target voxel and the map resolution to get the number and the location of adjacent voxels which met connectivity. Secondly, according to the different connectivity, the visibility model of the target voxel was built respectively to establish the full visibility model which was more universal. The proposed model could effectively overcome the limitations of the robust OctoMap visibility model, and improve the accuracy. Next, the simple depth error model was replaced by the Kinect sensor depth error model based on Gaussian mixture model to overcome the effect of the sensor measurement error on the accuracy of map further and reduce the uncertainty of the map. Finally, the Bayesian formula and linear interpolation algorithm were combined to update the occupancy probability of each node in the octree to build the volumetric occupancy map based on a octree. The experimental results show that the proposed method can effectively overcome the influence of Kinect sensor depth error on map precision and reduce the uncertainty of the map, and the accuracy of map is improved obviously compared with the robust OctoMap.

Key words: 3D map, robust OctoMap, Gaussian mixture model, octree, volumetric occupancy map

中图分类号: