Homography augmented particle filter SLAM

Abstract

The article presents a comprehensive study of a visual-inertial simultaneous localization and mapping (SLAM) algorithm designed for aerial vehicles. The goal of the research is to propose an improvement to the particle filter SLAM system that allows for more accurate and robust navigation of unknown environments. The authors introduce a modification that utilizes a homography matrix decomposition calculated from the camera frame-to-frame relationships. This procedure aims to refine the particle filter proposal distribution of the estimated robot state. In addition, the authors implement a mechanism of calculating a homography matrix from robot displacement, which is utilized to eliminate outliers in the frame-to-frame feature detection procedure. The algorithm is evaluated using simulation and real-world datasets, and the results show that the proposed improvements make the algorithm more accurate and robust. Specifically, the use of homography matrix decomposition allows the algorithm to be more efficient, with a smaller number of particles, without sacrificing accuracy. Furthermore, the incorporation of robot displacement information helps improve the accuracy of the feature detection procedure, leading to more reliable and consistent results. The article concludes with a discussion of the implemented and tested SLAM solution, highlighting its strengths and limitations. Overall, the proposed algorithm is a promising approach for achieving accurate and robust autonomous navigation of unknown environments.

Keywords

• simultaneous localization and mapping (slam)
• homography matrix
• particle filter
• robot navigation
• visual-inertial systems