High-precision 3D reconstruction with multi-line laser scanner based on SLAM framework

Abstract

This paper proposes a handheld multi-line laser 3D scanning system capable of real-time point cloud stitching. The hardware consists of dual industrial cameras and a laser projector, which can acquire laser line point clouds and marker point clouds from a single viewpoint. To realize the real-time fusion of multi-view point clouds, a high-precision point cloud stitching scheme is constructed based on simultaneous localization and mapping (SLAM) architecture. Specifically, an adaptive weighted iterative closest point (ICP) algorithm is adopted to match marker points for real-time handheld pose estimation; local mapping based on viewpoint classification is utilized to manage motion trajectories, and a multi-layer loop closure optimization structure is designed to suppress cumulative drift. Meanwhile, a stable landmark map is established to unify cross-view geometric constraints, which efficiently supports loop closure detection, viewpoint classification and relocalization, and improves the real-time performance of the system. Experimental results verify that the proposed method can achieve robust point cloud stitching without relying on two-dimensional features. Quantitative analysis shows that for locally layered point clouds, compared with the tracking-only scheme, the complete pipeline increases the volumetric density of processed local point clouds by 50% and reduces the surface thickness error by 33.04%. In addition, the scanning accuracy of the system remains stable within the range of 0.014-0.074 mm. The proposed scheme effectively addresses the key challenges of laser scanning, including insufficient stable features, inconsistent reconstruction errors and long-sequence pose drift, and balances the requirements of real-time performance and global high-precision stitching for multi-line laser reconstruction.