The paper analyzes the problems existing in the practical application of traditional laser triangulation algorithm and sums up the front-end and back-end of graph-based SLAM. Taking into account of the singularity, similarity and symmetry of garbage's feature which may cause feature mismatch and the shortcomings of reflective panels being easily blocked by other objects but with high position accuracy while triangulation was used, the idea of combining SLAM algorithm with triangulation algorithm is proposed. The differences between the traditional graph-based SLAM method and the improved one are compared. And the improved algorithm is applied into the navigation of overloaded AGV in the intelligent garage successfully.