一种融入注意力和预测的特征选择SLAM算法

PDF下载HTML

Title:

Feature selection simultaneous localization and mapping algorithm incorporating attention and anticipation

作者:

曾毓菁^1,2,3,4, 姜勇^2,3,4

1. 东北大学 信息科学与工程学院，辽宁 沈阳 110006;
2. 中国科学院 沈阳自动化研究所，辽宁 沈阳 110016;
3. 中国科学院 网络化控制系统重点实验室，辽宁 沈阳 110016;
4. 中国科学院 机器人与智能制造创新研究院，辽宁 沈阳 110169

Author(s):

ZENG Yujing^1,2,3,4, JIANG Yong^2,3,4

1. School of Information Science and Engineering, Northeastern University, Shenyang 110006, China;
2. Shenyang Institute of Automation, Chinese Academy of Sciences, Shenyang 110016, China;
3. Key Laboratory of Networked Control Systems, Chinese Acad

关键词:

即时定位与地图创建; 视觉; 注意力; 预测; 特征选择; logdet度量; 延迟求值; 贪婪算法; 信息矩阵

Keywords:

SLAM; vision; attention; anticipation; feature selection; logdet metric; lazy evaluation; greedy algorithm; information matrix

分类号:

TP391

DOI:

10.11992/tis.202010036

摘要:

针对SLAM (simultaneous localization and mapping)在急转弯、快速运动场景中定位失败的问题，提出一种融入注意力和预测的特征选择即时定位与地图创建(SLAM)算法，选择随着相机的运动更有可能保持在视野中的特征点，舍去即将消失在视野中的特征点。首先利用logdet度量量化特征选择的可行性，然后计算特征点的信息矩阵，再从检测到的特征中通过贪婪算法选择 k 个特征(近似的)最大化logdet度量，最后结合ORB-SLAM2的实际实验表明，该算法在复杂场景(如急转弯、快速运动)中可以确保定位的准确性。

Abstract:

A simultaneous localization and mapping (SLAM) algorithm incorporating attention and anticipation is proposed to solve the localization failure problem of SLAM in the scene of sharp turning and fast movement. The algorithm can select feature points that are more likely to remain in the field of view as the camera moves and discard features that are about to disappear from the field of view. The logdet metric is used to measure the feasibility of quantifying the feature selection first. The information matrix of the feature points is then calculated. From the detected features, a greedy algorithm is used to select k features (approximately) to maximize the logdet metric. The actual test combined with ORB-SLAM2 shows that the algorithm can ensure positioning accuracy in complex scenarios, such as in the scene of sharp turning and fast movement.