Główne pojęcia
This paper proposes an innovative continuous-time visual-inertial state estimation method based on Chebyshev polynomial optimization, which transforms the pose estimation problem into an optimization of polynomial coefficients and achieves higher accuracy compared to traditional preintegration methods.
Streszczenie
The paper presents a continuous-time visual-inertial state estimation algorithm based on Chebyshev polynomial optimization. The key highlights are:
-
Pose is modeled as a Chebyshev polynomial, with velocity and position obtained through analytical integration and differentiation. This transforms the continuous-time state estimation problem into a constant parameter optimization problem.
-
The optimization objective function incorporates the original IMU measurements, visual reprojection errors, and initial state constraints, avoiding the linearization issues in filtering methods and preserving the quasi-Gaussian nature of the measurements.
-
The use of Chebyshev polynomials ensures high accuracy and efficiency in the functional approximation. Simulation and experimental results on public datasets demonstrate that the proposed method outperforms traditional preintegration methods in both accuracy and computational efficiency.
-
The paper discusses the limitations of the current method, such as the lack of adaptive polynomial order selection and the focus on batch optimization. Future work will address these limitations by developing adaptive and real-time implementations of the Chebyshev polynomial optimization for visual-inertial state estimation.
Statystyki
The simulation results show that, compared to the preintegration method, the Chebyshev polynomial optimization achieves:
47% lower attitude accumulative RMSE, 58% lower velocity accumulative RMSE, and 65% lower position accumulative RMSE for the circular trajectory.
68% lower attitude accumulative RMSE, 49% lower velocity, and 59% lower position accumulative RMSE for the straight-line trajectory.
The experimental results on the EuRoC MAV dataset demonstrate that the Chebyshev polynomial optimization achieves:
30% improvement in velocity estimation accuracy and 50% improvement in position estimation accuracy on average.
50% improvement in computational efficiency on average.
Cytaty
"The core of VINS is the visual-inertial fusion state estimation algorithm."
"Optimization-based algorithms have sought to mitigate the errors induced by linearization."
"Continuous-time poses do not require the estimation of poses at each sensor measurement point, and the state dimension depends on the polynomial order of the pose representation, facilitating the fusion of sensors with different sampling rates."