Large-scale colored point clouds have many advantages in navigation or scene display. Relying on cameras and LiDARs, which are now widely used in reconstruction tasks, it is possible to obtain such colored point clouds. However, the information from these two kinds of sensors is not well fused in many existing frameworks, resulting in poor colorization results, thus resulting in inaccurate camera poses and damaged point colorization results. We propose a novel framework called Camera Pose Augmentation (CP+) to improve the camera poses and align them directly with the LiDAR-based point cloud. Initial coarse camera poses are given by LiDAR-Inertial or LiDAR-Inertial-Visual Odometry with approximate extrinsic parameters and time synchronization. The key steps to improve the alignment of the images consist of selecting a point cloud corresponding to a region of interest in each camera view, extracting reliable edge features from this point cloud, and deriving 2D-3D line correspondences which are used towards iterative minimization of the re-projection error.

翻译：摘要：大规模彩色点云在导航或场景展示方面具有诸多优势。借助当前广泛用于重建任务的相机与LiDAR传感器，可获取此类彩色点云。然而，现有许多框架未能有效融合这两类传感器的信息，导致着色效果不佳，进而造成相机位姿不准确与点云着色结果受损。本文提出一种名为“相机位姿增强”（CP+）的新框架，旨在优化相机位姿并使其直接与基于LiDAR的点云对齐。初始粗略相机位姿由LiDAR-惯性或LiDAR-惯性-视觉里程计结合近似外参参数与时间同步获得。优化图像对齐的关键步骤包括：为每个相机视野中的感兴趣区域选取对应点云，从该点云中提取可靠边缘特征，推导用于迭代最小化重投影误差的2D-3D线对应关系。