Abstract: With the growing demands of digitalization, data acquired from a single sensor has become inadequate for complex modeling tasks, and single-view approaches are inherently prone to data sparsity and occlusion issues. To address this problem, this paper proposes two sets of 3D reconstruction systems, which are equipped with multiple sensors including LiDAR, monocular cameras, and IMU, and can be respectively applied to multi-view data collection in both indoor and outdoor environments. Furthermore, this paper also presents a colorized 3D reconstruction technology based on the fusion of multi-source and multi-view data. The prerequisite is to obtain the coordinate system transformation relationships among various sensors through joint calibration. Then, the visual-inertial coupling system is utilized for motion estimation to acquire accurate postures and trajectories. Based on this, the single-view laser point clouds are optimized for distortion removal and assigned with true colors. Finally, multi-view point cloud stitching is carried out based on the mixed information of color and geometry to obtain the true-color 3D model of the scanned object. Tests and verifications have been conducted in both indoor and outdoor scenarios. The experimental results show that the 3D reconstruction accuracy of this method is higher than that of advanced algorithms, with the modeling error reaching the centimeter level. Additionally, the algorithm has higher robustness.

Key words: multi-source perception, multi-view point cloud matching, 3D reconstruction system, colorized 3D reconstruction