Efficient Minimal Solvers for Visual-Inertial Relative Pose Estimation in Multi-Camera Systems
tao li, zhenbao yu, Banglei Guan, jianli han, weimin lv
AI summary
Problem
Existing multi-camera relative pose estimators require excessive point correspondences or solve high-degree polynomials, hindering real-time performance and RANSAC efficiency.
Approach
By introducing a depth-based translation parameterization and leveraging IMU vertical or rotation axis priors, the authors derive two minimal 4-point solvers that reduce the pose estimation problem to solving a univariate 6th-degree polynomial.
Key results
- Two novel 4-point minimal solvers using IMU vertical and rotation axis priors
- Reduction of the generalized epipolar constraint to a 6th-degree univariate polynomial
- Superior computational efficiency and numerical stability over state-of-the-art solvers
- Competitive accuracy on synthetic data and the KITTI benchmark
Why it matters
Enables real-time visual odometry and SLAM for multi-camera systems in autonomous vehicles and UAVs by drastically reducing computational load and correspondence requirements.
Abstract
Estimating the relative poses of multi-camera systems is a fundamental problem in computer vision, with critical applications in autonomous vehicles, mobile devices, and unmanned aerial vehicles (UAVs). However, existing solutions often suffer from high computational complexity or rely on an excessive number of point correspondences, limiting their real- world applicability. To address these limitations, we propose two efficient minimal solvers for estimating the relative poses of multi-camera systems using a novel parameterization. The first solver leverages the vertical direction prior provided by Inertial Measurement Units (IMUs), while the second utilizes the rota- tion axis direction prior from IMUs. Our methods require only four point correspondences and reduce the problem of multi- camera relative pose estimation to solving a univariate 6th- degree polynomial—a significant improvement over existing ap- proaches, which typically involve 8th-degree polynomials. This reduction in computational complexity and correspondence requirements makes our solvers particularly effective when integrated into RANSAC frameworks, demonstrating strong potential for visual odometry applications. Through rigorous evaluations on synthetic data and the KITTI benchmark, our methods achieved superior computational efficiency and competitive accuracy compared to state-of-the-art algorithms.