A visual SLAM algorithm based on fuzzy clustering for removing dynamic features

Abstract

Most visual simultaneous localization and mapping (SLAM) algorithms assume that no or only few moving objects occur in application environments. This assumption makes the algorithms vulnerable to the interference of moving objects in dynamic environments. To address the problem, a new visual SLAM method, which could eliminate dynamic features without any prior information, was proposed. By measuring the position of each feature point and its motion vector difference between image sequences, a two-stage clustering was performed on the feature points in the field of view. This method removed the features detected on moving objects, and used a static initialization technique to eliminate the dependence of SLAM on prior information. The proposed method intended to improve OV2SLAM (a fully online and versatile visual SLAM for real-time applications) algorithm, and the experimental verification was carried out. Our results show that while maintaining the real-time performance of the original OV2SLAM algorithm, the positioning accuracy and robustness of the proposed method is improved in a dynamic environment.