We propose a general covariance estimation method for relative pose measurements using deep learning. Our approach extends previous system specific covariance estimation models. Such models map input images acquired from two different viewpoints to a covariance estimate. While such models have successfully been applied to relative pose measurements obtained from visual odometry, the extension to the general system scenario is rather more challenging. In this paper, we propose to map both the inputs images acquired from two viewpoints along with the relative pose measurement to a covariance estimate. By including the relative pose measurement as an additional input to the mapping, we show that it is possible to predict covariance for general relative pose measurements.
Deep Learning-Based Covariance Estimation for Relative Pose Measurements
2024-09-24
673450 byte
Conference paper
Electronic Resource
English
Optimal Pose Estimation with Error-Covariance Analysis
AIAA | 2021
|OPTIMAL POSE ESTIMATION WITH ERROR-COVARIANCE ANALYSIS
TIBKAT | 2021
|