Unified Pose Estimation for Multicamera Systems Using Point and Line Correspondences

Camera pose estimation from 2-D to 3-D point and line correspondences is a fundamental task in a wide range of computer vision applications. Although numerous methods have been proposed for single-camera systems, the multicamera case, which is known as the nonperspectiven- point-and-line (NPnPL) problem, remains largely underexplored. Existing solutions are primarily minimal solvers tailored for specific configurations, lacking a general framework to handle arbitrary numbers of correspondences efficiently. In this article, we propose UPnPL, a unified and efficient solver for the NPnPL problem. Our approach introduces a general geometric formulation applicable to arbitrary camera configurations, building upon an O(n) EPnP-based framework. To maintain linear complexity, we use endpoints to establish linear constraints for 3-D lines. However, since endpoint ambiguity can degrade accuracy, we further introduce a coordinate-invariant robust selection strategy. The final pose is recovered through a resultant-based polynomial solver. Extensive experiments on both synthetic and realworld datasets, including EuRoC and KITTI-360, demonstrate that our method provides a general and state-of-the-art solution for multicamera localization using point and line features. The code is publicly available https://github.com/xiaoli-1664/upnpl.git