Visual SLAM Based on Geometric Cluster Matching

• Published in: 2019 IEEE International Conference on Real-time Computing and Robotics (RCAR) pp. 923 - 928
• Main Authors: Tian, Chaoran, Ou, Yongsheng, Qu, Yangyang
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01-08-2019
• Subjects: Cameras; Clustering algorithms; Feature extraction; Impedance matching; Simultaneous localization and mapping; Visualization
• DOI: 10.1109/RCAR47638.2019.9044135

The approach of feature-based visual SLAM is very popular in Micro Aerial Vehicle (MAV) navigation and wheeled mobile robot (WMR) navigation. Most modern feature-based visual SLAM frameworks extract a fixed number of feature points and then match the feature points by brute force or adopt Fast Library for Approximate Nearest Neighbors (FLANN). The features' extracting and matching processes will determine the localization accuracy and mapping quality significantly. In this paper, we propose a feature matching algorithm for the robot equipped with a binocular camera, which provides accurate real-time localization performance and robustness. In our feature matching method, both the geometric relationship of image feature points and the similarity of descriptor are considered. This paper presents a visual SLAM framework based on geometric cluster matching and the effectiveness of the proposed method is verified by practical experiments.