Multiview Minimal Efficient Solution of Point–Line Camera Pose

Pose estimation involves the problem of determining the orientation and position of the calibrated camera by known 3-D world features (i.e., points and lines) and their 2-D image projections. While the prior method by Miraldo et al. (2018) has explored scenarios involving both points and lines, they primarily focused on the generic case and neglected the coplanar situation. In this article, we derive a two-point and line-based minimal pose method (MV-P2P1L) in the generic and coplanar cases, which transitions from the state-of-the-art single-view solution by Hruby et al. (2024) to the multiview minimal solution. This method projects point and line features into different calibrated cameras, respectively. By utilizing calibrated cameras with 3-D features and 2-D projections, the minimal pose problem easily involves computing the roots of a fourth-degree equation in the generic case or a second-degree equation in the coplanar case. Moreover, the proposed pose solution is combined with the random sample consensus (RANSAC) algorithm to effectively reject feature outliers. We experimentally demonstrate that the proposed MV-P2P1L achieves higher accuracy and comparable performance to existing state-of-the-art solutions in runtime.