Aiming to address the degradation of robot mapping and localization accuracy caused by complex terrain and high similarity in unstructured environments, a SLAM algorithm based on multisensor loop detection, namely multi-sensor loop detection LiDAR-IMU LOAM ( MLD-LOAM) was proposed. Based on the LEGO LOAM algorithm architecture, this algorithm incorporated IMU data in the ground point cloud filtering phase after rapid filtering based on the distance distribution of laser points. This algorithm then performed a more accurate secondary filtering of the ground point cloud by calculating ground normals in real time. In the loop detection phase,the IMU and lidar data were fused by using a LiDAR confidence function designed with the sparsity of laser point cloud features. The fused odometry data triggered loop detection, which was then performed by an octree-modified normal distribution transform ( NDT), thereby improving system accuracy and efficiency. Experiments were conducted on the KITTI dataset and in unstructured outdoor real-world scenarios. Results showed that compared with LEGO-LOAM, MLD-LOAM improved localization accuracy by 11% and 30% in both the KITTI dataset and real-world scenarios. Compared with LIO-SAM and FAST-LIO, MLD-LOAM achieved better localization accuracy in real-world scenarios than FAST-LIO, but it was 2.8% less accurate than LIO-SAM. However, the speed of its memory consumption was only 50% of LIO-SAM and 32.4% of FAST LIO. The research result can provide a feasible solution for long-term mapping and localization tasks of outdoor robots that required low memory footprint.