A Closed-loop Detection Algorithm Based on Dynamic Time Warping

In the research of closed-loop detection, algorithms based on single image matching are drawing great research interests. However, in dynamic environment, such algorithms can not meet the requirements of dynamic detection.Considering that a single image is correlated in space and time, this paper pr...

Full description

Saved in:
Bibliographic Details
Main Authors: ZHU Tian, PENG Zhichuan, ZHANG Zhiteng, ZHU Zemin, XIE Yongbo, WANG Wenming
Format: Article
Language:zho
Published: Editorial Office of Control and Information Technology 2021-01-01
Series:Kongzhi Yu Xinxi Jishu
Subjects:
Online Access:http://ctet.csrzic.com/thesisDetails#10.13889/j.issn.2096-5427.2021.04.014
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:In the research of closed-loop detection, algorithms based on single image matching are drawing great research interests. However, in dynamic environment, such algorithms can not meet the requirements of dynamic detection.Considering that a single image is correlated in space and time, this paper proposes a closed-loop detection algorithm based on dynamic time warping.This method solves the similarity of image sequences by dynamic time-warping algorithm, instead of solving the similarity between single images. At the same time, it uses convolutional neural network to extract image features and the results are more advanced and abstract, and this enhances the robustness of the algorithm. Considering the influence of matching sequence length on the experimental results, the quasi-call curves and ROC curves obtained under different sequence lengths are compared in detail to determine the most suitable matching sequence length. Experiments show that the loop closure detection algorithm based on dynamic time warping algorithm has higher accuracy and recall rate, and the false positive probability is lower. Compared with the mainstream methods such as FabMap and SeqSLAM, the performance of the loop closure detection algorithm is better.
ISSN:2096-5427