PVE-LIOM: Pseudo-Visual Enhanced LiDAR-Inertial Odometry and Mapping

Traditional light detection and ranging (LiDAR) localization and mapping approaches for solid-state LiDAR may experience performance degradation when operating in buildings and may struggle to detect reliable loop closures. Recently, there has been a growing interest in the use of visual-aided LiDAR for localization and mapping. Motivated by this, this article proposes a pseudo-visual enhanced LiDAR-inertial odometry and mapping (PVE-LIOM) framework for solid-state LiDAR that utilizes reflectivity information from LiDAR to generate pseudo-images instead of relying on real camera images. Although the information in the pseudo-images may not be as rich as that obtained from real camera images, they offer the advantage of remaining stable under dynamic illumination. By utilizing pseudo-images, this framework introduces an adaptive voxel filter, an automatic failure detection module and a loop detection module, which enhance the robustness of odometry estimation with lower cumulative error in the front-end and improve the reliability of loop closure detection with better pose-graph optimization (PGO) in the back-end. Notably, the pseudo-vision-based back-end module is designed to be plug-and-play. Our framework is evaluated in various environments and compared with state-of-the-art (SOTA) algorithms. The results demonstrate superior performance in terms of both less cumulative drift and excellent global map consistency.