Abstract: In an outdoor environment with too many features in a large area, problems like accuracy errors and lack of robustness of a simultaneous localization and mapping (SLAM) system with multiple sensors may occur because of feature mismatch. In this view, this paper proposes an improved voxelized generalized iterative closest point (VGICP) method. Firstly, a feature credibility screening method is proposed to provide the system with an accurate initial guess by using the complementary characteristics of laser-inertial navigation-vision sensors to perceive different environments. Then, the visual feature subset is associated with the point cloud data through depth information, and the voxelized target point cloud group with high observability is screened by adding visual constraints, which makes the positioning and mapping more accurate while reducing the computational complexity. The simulation experiments show that, when the SLAM system based on multi-sensor fusion builds maps in an environment with more feature point clouds, the positioning accuracy improves by 12.335% compared with that of LVI-SAM system. When the operation linear speed exceeds 10 m/s, the robustness of the system is improved, which has strong feasibility.