APP下载

基于thLSH的快速闭环检测方法

2018-09-29姜海洋朴松昊张祖亮

智能计算机与应用 2018年4期

姜海洋 朴松昊 张祖亮

摘 要:同步定位與地图构建(SLAM)问题是实现移动机器人在未知环境中自我定位和导航的关键技术,具有重大的理论意义和研究价值。闭环检测是SLAM中的一个关键模块,对机器人实时更新地图和避免引入错误的地图节点起着关键作用。本文将视觉闭环检测问题看作是图像检索问题,基于深度学习的思想,将RandNet神经网络应用到闭环检测,并利用基于阈值的局部敏感哈希算法对其提取的图像特征匹配过程进行加速。实验结果表明:本文所提出的基于thLSH的快速闭环检测方法,在保证高准确率的前提下,特征匹配速度提升了10倍,能够更好满足闭环检测的实时性需求。

关键词:同步定位和地图构建; 闭环检测; 神经网络; 局部敏感哈希

Abstract: The issue of Simultaneous Localization and Mapping (SLAM) is a crucial technology for autonomous positioning and navigation of mobile robots in an unknown environment, and has theoretical significance and research value. Loop closure detection is a vital module in SLAM and plays an important role in updating the map in real time and avoiding the introduction of wrong map nodes. The loop detection vision problem in this thesis is regarded as an image retrieval issue, and based on the idea of deep learning, the RandNet neural network is applied to the loop detection problem, and a threshold-based locality-sensitive hashing algorithm is also used to accelerate the image feature matching process. The experimental results show that the fast loop detection method based on thLSH proposed in this paper can improve the feature extraction speed by more than 3 times on the premise of high accuracy, the speed of feature matching is 10 times higher, and it can meet the real-time requirement of loop detection.

Key words: visual Simultaneous Localization and Mapping; loop closure detection; neural network; locality-sensitive hashing

引言

同步定位和建图问题(simultaneous localization and mapping,简称SLAM)中的即时定位和建图,是利用算法对传感器在未知环境中采集的信息进行处理,由于环境的未知性、建模算法本身的缺陷、传感器的累计误差等问题,致使SLAM的定位与建图的精度不够准确,甚至可能导致环路不能闭合,而无法判断机器人是否回到已访问过的区域。SLAM中的闭环检测(Loop Closure Detetion,LCD)模块是通过判断当前位置是否处于已访问过的区域,抑制传感器的累积误差[1],进而提高后端图优化的准确性,保证地图构建的效果的精准表达。

闭环检测,本质上是数据的关联问题。在基于视觉同步定位与地图构建(visual Simultaneous Localization and Mapping, vSLAM)系统中,通常采用的方法是将当前位置采集的场景图像与此前所采集的场景图像进行匹配,当2个场景图像的相似度达到一定阈值时,即可判定发生了闭环。……

登录APP查看全文