In this paper, we present an iterative two-stage scheme for precise and robust frame-to-frame feature-based ego-motion estimation using stereo cameras. We analyze the characteristics of the optical flows and reprojection errors that are independently induced by each of the decoupled six degrees of freedom motion. As we will show, the different characteristics of these induced optical flows lead to a reprojection error that depends on the coordinates of the features. When using a proper normalization of the reprojection error, this coordinate-dependency can be almost completely removed for decoupled motions. Furthermore, we present a way to use these results for automotive application where rotation and forward motion are coupled. This is done by compensating for the flow that is induced by the rotation, which decouples the translation flow from the overall flow. The resulting method generalizes the ROCC approach [4], where a robust outlier criterion was introduced and proved to increase robustness and quality for large forward translation motions. Therewith the proposed method generalizes ROCC to almost all possible automotive motions. The performance of the method is evaluated on Kitti benchmark and currently2 reaches the best translation error of all camera-based methods.
Flow-decoupled normalized reprojection error for visual odometry
2016-11-01
2090870 byte
Conference paper
Electronic Resource
English
DeLiO: Decoupled LiDAR Odometry
IEEE | 2019
|British Library Conference Proceedings | 2004
|IEEE | 2004
|Shape from incomplete silhouettes based on the reprojection error
British Library Online Contents | 2010
|