Three-dimensional lidar is widely used in simultaneous localization and mapping (SLAM) research due to its accurate and reliable measurement performance, and has achieved fruitful results. However, in the scene where the geometric features such as long corridors and tunnels are not rich enough, the point cloud registration will have an additional degree of freedom in one direction, and the laser SLAM based on the point cloud information will degenerate, resulting in the failure of robot positioning and mapping, which will lead to the failure of subsequent navigation tasks. That is to say, when moving along a long corridor, the laser point cloud obtained is the same, which makes the matching algorithm unable to accurately estimate the motion in this direction. And in the conventional environment, the accumulated error of the front-end odometer increases with the increase of the scene. The loopback detection method based on the Euclidean distance of the traditional laser SLAM algorithm has the problem of missing detection and cannot eliminate the accumulated error of the front-end odometer. For this reason, a close-coupled laser SLAM method for an unknown environment is proposed in this paper.First of all, a close-coupled framework (Fig.1) is used to fuse LiDAR and IMU information, correct IMU bias, and provide high-precision prior information for the LiDAR odometer. Secondly, the LiDAR odometer Jacobi matrix is calculated, the environmental geometric information dimension is detected in real-time, the wheel odometer and IMU data is integrated, and the freedom of LiDAR odometer is compensated. Finally, in view of the loopback detection method based on Euclidean distance that has missed detection due to the accumulated error of the odometer (Fig.2), a variable-threshold loopback search model is constructed, and the corresponding threshold key frame information is analyzed by different registration methods to improve the loopback detection recall rate (Fig.3).A-LOAM, LIO-SAM and the method in this paper are used to test in a 76.78 m long and 1.85 m wide corridor respectively. The corridor is flanked by large white walls, with high scene repetition and fewer geometric features in the direction of motion. In the long corridor degradation scenario, the scene restoration degree of the method in this paper reaches 99.71%, and the odometer drift is reduced to 0.12 m, which is 91.04% and 97.37% lower than A-LOAM and LIO-SAM, respectively (Tab.1). In addition, the data sets with three drift thresholds are selected, and 20 data sets are selected for each threshold to test the loopback detection performance. Under the condition that the accuracy of loopback detection is 100%, the recall rate of loopback detection is 98.3%, which is 35% higher than that of the LIO-SAM algorithm on average (Tab.3).In this study, a close-coupled laser SLAM method for an unknown environment is proposed. This method uses a tight coupling framework to improve the efficiency of sensor information utilization, and can detect the geometric information dimension of the environment in real-time. In the environment of fast-rotating scenes and missing geometric information, it can still achieve high-precision positioning and mapping. A variable threshold loopback search model is builded, which can maintain a high loopback detection recall rate in large scenes. The position and posture of historical frames are corrected, and the global consistency of the map is ensured. The proposed method is proved to be robust and accurate by many scene experiments.