Abstract: Aiming at the unstructured degradation environment of underground mines，the laser point cloud map generated by conventional laser-inertial navigation SLAM technology has problems such as lack of texture information，poor visualization effect，and insufficient absolute accuracy under the interference of dynamic objects. It is difficult to meet the engineering requirements of fine acceptance and safety risk assess⁃ ment of roadways. A SLAM acquisition system integrating lidar，industrial camera and inertial navigation sensor is studied and developed. The hardware integration design and space-time synchronization scheme are completed. The FAST-LIVO2 tightly coupled SLAM algorithm is improved，and the M-detector dynam⁃ ic target detection module is embedded. The forward mutual promotion mechanism of dynamic point filtering and pose optimization is established. Based on the 280 m production roadway of an underground mine in Gui⁃ zhou，the field test was carried out，and the coordinate transformation and accuracy evaluation were carried out with the measured target center coordinates of the total station as the true value. The results show that the absolute mapping accuracy of the system can reach 13.5 cm in the range of 280 m roadway. Compared with the non-fusion dynamic object detection，the mapping accuracy is improved by more than 20 %. The true color point cloud map generated has clear texture，good color restoration，and high composite degree with roadway CAD design map. Finally，based on the true color point cloud data，the roadway safety risk assess⁃ ment is carried out，which verifies the practical value of the system in the digitization and safety manage⁃ ment of underground mines，and provides technical support for the fine perception of three-dimensional space in roadways.

Key words: underground mines, laser-vision-intertial fusion, SLAM, true color point cloud, absolute mapping accuracy, safety risk assessment