文章基本信息

标题：FIXED PARTITIONING USING AN HEURISTIC APPROACH FOR LOOP CLOSURE DETECTION
本地全文：下载
作者：AZIZI ABDULLAH ; IZWAN AZMI ; MOHAMMAD FAIDZUL NASRUDIN 等
期刊名称：Journal of Theoretical and Applied Information Technology
印刷版ISSN：1992-8645
电子版ISSN：1817-3195
出版年度：2021
卷号：99
期号：8
语种：English
出版社：Journal of Theoretical and Applied
摘要：Loop closure detection using visual information needs proper representation to interpret objects in a scene effectively. The scene may contain several objects corresponding to known and new landmarks. The most widely used methods to detect and describe the regions is to use a keypoint detector to localize objects such as in speeded-up robust features (SURF). Most keypoint-based schemes compute salient points on the object based on the curvature principles of geometrical object surfaces. However, not all the object surfaces can distinctively be described using these rules, such as in scenery images that contain high repeating texture features. In the keypoint detector scheme does not consider the flat texture regions resulting in a few detected points which hinder the holistic visual description of images. Thus, we propose to use a fixed-partitioning scheme that divides the image into several blocks for grouping spatial semantic of significance image features. One possible problem in the proposed approach is to identify the number of partitions and partition size for image description. Thus, an heuristic approach is used to identify these parameters for loop closure detection. A famous computational expensive Real-Time Appearance Based Mapping (RTAB-Map) simultaneous localization and mapping (SLAM) is used to validate the proposed scheme. The results show that the proposed approach outperforms the standard keypoint detector on two datasets, namely Lib6 Indoor and New College.
关键词：Vision-based SLAM;Loop-closure detection;localization;Fixed Partitioning