Fig 1.
Overall architecture diagram.
Fig 2.
Schematic diagram of point-to-plane association strategy.
Fig 3.
Flowchart of plane feature construction and matching.
Fig 4.
Flowchart of posture decoupling based on Manhattan coordinate system.
Fig 5.
The specific definitions of MF and MP, as well as the relationship between them.
Fig 6.
The relationship between the world coordinate system (W), the historical frame camera coordinate system (L), the Manhattan coordinate system (M), and the current frame camera coordinate system (C).
Fig 7.
Two parallel orthogonal plane features.
Fig 8.
One parallel or orthogonal plane feature and another orthogonal vertical plane feature.
Fig 9.
Decoupling of the translation matrix.
Fig 10.
Pseudocode implementation of pose matrix decoupling.
Fig 11.
Constructing imaginary plane features using common points in frame.
Fig 12.
Constructing imaginary plane features using common points in frame.
Fig 13.
Pose estimation model based on joint optimization of points, lines and planes and vanishing point constraints.
Fig 14.
Scene of long corridor.
Fig 15.
Line feature classification based on vanishing points.
Fig 16.
Iteration between Manhattan coordinate system and vanishing point.
Table 1.
The module comparison of related classical algorithms.
Table 2.
Evaluation results of translation ATE RMSE (unit: m) on ICL-NUIM and TUM datasets. Bold numbers represent the best performances. ‘×’ represents tracking lost.
Fig 17.
RMSE comparison of different algorithms in the ICL dataset living_room_traj1_frei.
Fig 18.
RMSE comparison of different algorithms in the ICL dataset living_room_traj2_frei.
Fig 19.
Trajectory comparison of different algorithms in the ICL dataset living_room_traj1_frei.
Fig 20.
Trajectory comparison of different algorithms in the ICL dataset living_room_traj2_frei.
Fig 21.
Trajectory comparison of different algorithms in the TUM dataset 2_desk_with_person.
Fig 22.
Trajectory comparison of different algorithms in the TUM dataset 2_xyz.
Fig 23.
Trajectory comparison of different algorithms in the TUM dataset 3_long_office_household.
Fig 24.
Trajectory comparison of different algorithms in the TUM dataset 3_structure_texture_near.
Fig 25.
RMSE comparison of different algorithms in the TUM dataset 2_desk_with_person.
Fig 26.
RMSE comparison of different algorithms s in the TUM dataset 2_xyz.
Fig 27.
RMSE comparison of different algorithms in the TUM dataset 3_long_office_household.
Fig 28.
RMSE comparison of different algorithms in the TUM dataset 3_structure_texture_near.
Table 3.
Frame counts for MF extraction in the experimental datasets.
Fig 29.
Box plot of ATE of 1_rpy for different algorithms.
Fig 30.
Box plot of ATE of living_room_traj1_frei for different algorithms.
Fig 31.
Average ATE Comparison of SLAM Algorithms on TUM and ICL Datasets.
Fig 32.
Comparison of the number of plane feature extractions between scenarios in structure_texture_far.
Fig 33.
Comparison of the number of plane feature extractions between scenarios in structure_texture_near.
Fig 34.
Extraction of line features in the structure_texture_far sequence.
Fig 35.
Classification based on vanishing points in the structure_texture_far sequence.
Fig 36.
Orthogonal plane features extracted from the current frame in the structure_texture_far sequence.
Fig 37.
Orthogonal plane features extracted from orthogonal plane features matched in the historical frame in the structure_texture_far sequence.
Fig 38.
ZED2i camera-based test equipment.
Fig 39.
Scene of long corridor scene.
Fig 40.
Scene of underground parking scene.
Table 4.
Evaluation results of translation ATE RMSE (unit: m) on long corridor scene and the underground garage scene. Bold numbers represent the best performances.
Fig 41.
Trajectory comparison of Data_1-1 based on the actual scene measured by ZED2i camera.
Fig 42.
Trajectory comparison of Data_2-2 based on the actual scene measured by ZED2i camera.
Fig 43.
Trajectory comparison of Data_2-3 based on the actual scene measured by ZED2i camera.
Fig 44.
RMSE comparison of Data_1-1 based on the actual scene measured by ZED2i camera.
Fig 45.
RMSE comparison of Data_2-2 based on the actual scene measured by ZED2i camera.
Fig 46.
RMSE comparison of Data_2-3 based on the actual scene measured by ZED2i camera.
Fig 47.
Comparison of the number of extracted plane features in scenes Data_1-1.
Fig 48.
Comparison of the number of extracted plane features in scenes Data_1-2.
Fig 49.
Comparison of the number of extracted plane features in scenes Data_2-1.
Fig 50.
Comparison of the number of extracted plane features in scenes Data_2-2.