Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

An Improved Multi-State Constraint Kalman Filter for Visual-Inertial Odometry

M. R. Abdollahi, Seid H. Pourtakdoust, M. H. Yoosefian Nooshabadi, H. N. Pishkenari

TL;DR

It is demonstrated that the proposed Fast-MSCKF (FMSCKF) is about six times faster and at least 20% more accurate in final position estimation than the standard MSCKF algorithm.

Abstract

Fast pose estimation (PE) is of vital importance for successful mission performance of agile autonomous robots. Global Positioning Systems such as GPS and GNSS have been typically used in fusion with Inertial Navigation Systems (INS) for PE. However, the low update rate and lack of proper signals make their utility impractical for indoor and urban applications. On the other hand, Visual-Inertial Odometry (VIO) is gaining popularity as a practical alternative for GNSS/INS systems in GPS-denied environments. Among the many VIO-based methods, the Multi-State Constraint Kalman Filter (MSCKF) has received a greater attention due to its robustness, speed and accuracy. To this end, the high computational cost associated with image processing for real-time implementation of MSCKF on resource-constrained vehicles is still a challenging ongoing research. In this paper, an enhanced version of the MSCKF is proposed. To this aim, different feature marginalization and state pruning strategies are suggested that result in a much faster algorithm. The proposed algorithm is tested both on an open-source dataset and in real-world experiments for validation. It is demonstrated that the proposed Fast-MSCKF (FMSCKF) is about six times faster and at least 20% more accurate in final position estimation than the standard MSCKF algorithm.

An Improved Multi-State Constraint Kalman Filter for Visual-Inertial Odometry

TL;DR

It is demonstrated that the proposed Fast-MSCKF (FMSCKF) is about six times faster and at least 20% more accurate in final position estimation than the standard MSCKF algorithm.

Abstract

Fast pose estimation (PE) is of vital importance for successful mission performance of agile autonomous robots. Global Positioning Systems such as GPS and GNSS have been typically used in fusion with Inertial Navigation Systems (INS) for PE. However, the low update rate and lack of proper signals make their utility impractical for indoor and urban applications. On the other hand, Visual-Inertial Odometry (VIO) is gaining popularity as a practical alternative for GNSS/INS systems in GPS-denied environments. Among the many VIO-based methods, the Multi-State Constraint Kalman Filter (MSCKF) has received a greater attention due to its robustness, speed and accuracy. To this end, the high computational cost associated with image processing for real-time implementation of MSCKF on resource-constrained vehicles is still a challenging ongoing research. In this paper, an enhanced version of the MSCKF is proposed. To this aim, different feature marginalization and state pruning strategies are suggested that result in a much faster algorithm. The proposed algorithm is tested both on an open-source dataset and in real-world experiments for validation. It is demonstrated that the proposed Fast-MSCKF (FMSCKF) is about six times faster and at least 20% more accurate in final position estimation than the standard MSCKF algorithm.
Paper Structure (14 sections, 53 equations, 22 figures, 6 tables, 1 algorithm)

This paper contains 14 sections, 53 equations, 22 figures, 6 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. The Multi State Constraint Kalman Filter (MSCKF)
  3. The Propagation Step
  4. The Augmentation Step
  5. The Update Step
  6. Feature Marginalization And State Pruning
  7. Results
  8. Public Dataset
  9. Real-world Experiments
  10. Short-range experiment
  11. Mid-range experiment
  12. Long-range experiment
  13. Discussion
  14. Conclusion

Figures (22)

  • Figure 1: The flowchart of feature management approaches in the MSCKF and the FMSCKF algorithms.
  • Figure 2: Comparison of the number of extracted features in the FMSCKF and the MSCKF. In the fourth frame on the left column, the number of tracked features falls below the threshold, and therefore, the algorithm extracts new features. However, the number of features in the original MSCKF algorithm, shown in the right column, is constant. The images are parts of MH_$01$ dataset ref27.
  • Figure 3: orientation estimation RMSE for different algorithms.
  • Figure 4: position estimation RMSE for different algorithms.
  • Figure 5: Estimated 3D paths for MH_$01$ using the MSCKF (blue) and the FMSCKF (red). The ground-truth path is plotted in green ref27.
  • ...and 17 more figures