A Compact Loop Closure Detection Based on Spatial Partitioning

被引：0

作者：

Chen, Jianbin ^{[1
]}

Li, Jun ^{[1
]}

Xu, Yang ^{[1
]}

Shen, Guangtian ^{[1
]}

Gao, Yangjian ^{[1
]}

机构：

[1] Chongqing Univ, Automat Coll, Chongqing, Peoples R China

来源：

2017 2ND INTERNATIONAL CONFERENCE ON IMAGE, VISION AND COMPUTING (ICIVC 2017) | 2017年

基金：

中国国家自然科学基金;

关键词：

Loop closure detection; BoW; K-mean; Scene segmentation; BAGS;

D O I：

暂无

中图分类号：

TP18 [人工智能理论];

学科分类号：

081104 ; 0812 ; 0835 ; 1405 ;

摘要：

Loop closure detection (LCD) is a process trying to find a match between the current and a previously visited locations in SLAM. The bag of words (BoW) is a popular approach used in LCD, however, limited by perceptual aliasing primarily due to vector quantization. This paper proposes an improved method of the BoW called spatial partitioning BoW(SPBoW). We first apply scene segmentation to integrate the spatial information of visual features into BoW. Then, for better error tolerance, we setup a hierarchical K-means association dictionary to relate all visual words. Finally, sliding optimization is used to eliminate the effect of changing perspectives. The experiment results demonstrate much better real-time reaction and recall performance compared with the conventional BoW algorithm.

引用

页码：371 / 375

页数：5