Visual odometry improves by modeling uncertain feature matches spatially
MomentBA: Second-order Spatial Moments for Anisotropic Correspondence Uncertainty in Differentiable Bundle Adjustment
Computer Vision and Pattern Recognition
Summary
Visual odometry systems estimate motion by matching points between camera images. Usually, these matches are treated as exact or assumed equally uncertain, but in reality some matches are less reliable, especially along edges or blurry areas. The authors propose a method, MomentBA, that calculates detailed uncertainty from the shape of matching signals, which helps the system know which points to trust more. They show this improves motion estimates in test environments, especially reducing rotation errors.
What this means in practice
- •For robotics engineers: Enhance robot navigation by using spatial uncertainty estimates to improve camera-based motion tracking accuracy in challenging visual environments.
- •For augmented reality developers: Improve the stability and precision of AR device pose estimation under difficult conditions such as motion blur or repetitive textures by integrating anisotropic uncertainty modeling.
Authors
Yuqing Wang, Xiaoji Niu, Yan Wang, Hailiang Tang, Jian Kuang, Tisheng Zhang
Abstract
Most existing visual odometry (VO) systems treat feature correspondences as deterministic measurements or assign uniform uncertainty, ignoring the inherent localization ambiguity of different observations. However, correspondence uncertainty is often anisotropic due to image structures such as edges, repetitive patterns, and motion blur, which can significantly affect geometric optimization. In this work, we propose MomentBA, a geometry-aware bundle adjustment framework that derives anisotropic correspondence uncertainty from second-order spatial moments of local similarity responses. Instead of introducing additional covariance prediction networks, the proposed method directly converts matching response distributions into interpretable covariance estimates and incorporates them into bundle adjustment as correspondence-specific information matrices for uncertainty-aware residual weighting. Furthermore, the proposed formulation is integrated into a differentiable optimization framework, establishing a direct connection between correspondence uncertainty and geometric estimation. Experiments on the EuRoC MAV and TartanAir v1 Hard datasets demonstrate that MomentBA improves monocular visual odometry accuracy compared with existing feature-based and learning-based approaches. The proposed anisotropic covariance model achieves lower rotational errors and more robust trajectory estimation than fixed and isotropic uncertainty models, validating the effectiveness of geometry-induced uncertainty modeling for challenging visual environments.