Bayesian method improves homography estimates with uncertainty
Closed-form Bayesian homography estimation from noisy point correspondences
Computer Vision and Pattern Recognition
Summary
Estimating how one image maps onto another is crucial in computer vision but is often uncertain due to noisy data points. The authors propose a new Bayesian method that not only finds the best mapping but also quantifies the uncertainty of that estimate. This approach improves accuracy compared to traditional methods and works well on both simulated and real images. The added uncertainty information can help make downstream tasks more reliable.
What this means in practice
- •For computer vision engineers: Improve image stitching reliability by using homography estimates that include uncertainty information.
- •For autonomous vehicle developers: Enhance vehicle perception systems by integrating homography with quantified uncertainty for safer environment mapping.
Authors
Hanne Beuter, Sebastian Dorn
Abstract
While homographies are fundamental to many computer vision tasks, the majority of conventional estimation techniques provide only point estimates without directly quantifying uncertainty introduced by noisy observations. Uncertainty, though, propagates to subsequent processing steps such as camera calibration and 3D reconstruction and is particularly relevant in safety-critical and socially relevant fields including medical imaging, autonomous driving, and defense. We present a fast Bayesian formulation for homography estimation from point correspondences that explicitly incorporates measurement uncertainty and prior knowledge while providing a posterior distribution over the homography parameters. A closed-form solution of the posterior mean of the homography is derived in homogeneous coordinates and supplemented by an iterative Bayesian approach to handle non-linearities. Synthetic experiments demonstrate the applicability to projective transformations and show improved estimation accuracy over DLT under varying noise conditions. Image stitching experiments further demonstrate applicability to real image correspondences while additionally providing uncertainty information.