INTEGRATING CNN AND DICTIONARY MECHANISMS FOR EFFECTIVE LOOP CLOSURE DETECTION

Loop closure detection (LCD) remains a critical challenge in visual Simultaneous Localization and Mapping (SLAM), particularly in environments with repetitive structures or sparse textures where traditional methods suffer from perceptual aliasing and computational inefficiency. This paper presents...

Full description

Saved in:
Bibliographic Details
Main Authors: Ayda Mohammed Sharif, Sadegh Abdollah Aminifar
Format: Article
Language:English
Published: University of Zakho 2025-07-01
Series:Science Journal of University of Zakho
Subjects:
Online Access:https://sjuoz.uoz.edu.krd/index.php/sjuoz/article/view/1579
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Loop closure detection (LCD) remains a critical challenge in visual Simultaneous Localization and Mapping (SLAM), particularly in environments with repetitive structures or sparse textures where traditional methods suffer from perceptual aliasing and computational inefficiency. This paper presents a robust and scalable LCD framework that integrates a lightweight Convolutional Neural Network (CNN) with a dictionary-based voting mechanism, optimized for accuracy and real-time performance in resource-constrained settings. The proposed CNN architecture, featuring a single convolutional layer with 32 filters, achieves 98% classification accuracy on the Greenhouse Scene Dataset-a structured agricultural environment. Complementing the CNN, a dynamic dictionary tracks class frequencies to detect loop closures via adaptive thresholding, eliminating the need for complex feature matching or geometric verification. Experimental results demonstrate real-time operation (0.076 seconds per 70 frames) and resilience to spatial distortions, maintaining 92% accuracy under pixel-level shifts. Compared to state-of-the-art methods, our approach reduces computational overhead and memory usage
ISSN:2663-628X
2663-6298