Author Affiliations
Institute of War Support, Rocket Force University of Engineering, Xi'an, Shaanxi 710025, Chinashow less
Fig. 1. Flowchart of monocular visual odometry
Fig. 2. Feature extraction of points (green) and lines (red) in Tum dataset
Fig. 3. Schematic of error analysis for different initialization models
Fig. 4. Schematic of depth filter
Fig. 5. Real-time operation effects in part of the Euroc datasets. (a) MH_01 dataset; (b) MH_02 dataset
Fig. 6. Location effect in partial Euroc datasets. (a) MH_01 dataset; (b) MH_02 dataset
Fig. 7. Real-time operation effects in part of the Tum datasets. (a) fr2_desk dataset; (b) fr2_xyz dataset
Fig. 8. Location effects in partial Euroc datasets. (a) fr2_desk dataset; (b) fr2_xyz dataset
Fig. 9. Physical map of robot platform
Fig. 10. Running effects of actual robot. (a) Indoor scene; (b) corridor scene
Experiment 1 | Timeconsuming /ms | Error /pixel |
---|
Line featurealignment | 1.737 | 0.0059 | Point featureallignment | 1.815 | 1.2912 |
|
Table 1. Comparison of correction performance for point-line features
Experiment 2 | Timeconsuming /ms | Convergencenumber | Averageerror ofdepth value /cm |
---|
Depthestimation ofline feature | 215.33 | 56 | 18.25 | Depthestimation ofpoint feature | 277.98 | 286 | 15.89 | Depthestimation ofpoint-linefeature | 347.97 | 342 | 16.27 |
|
Table 2. Comparison of depth estimation performance for point-line features
Experiment 3 | RMSE /m |
---|
fc_mono | SVO | LSD-SLAM(no loop) |
---|
MH_01 | 0.16 | 0.17 | 0.18 | MH_02 | 0.21 | 0.27 | 0.56 | MH_03 | 1.70 | 0.43 | 2.69 | MH_04 | 2.71 | 1.36 | 2.13 | Vicon1_01 | 0.82 | 0.20 | 1.24 |
|
Table 3. Error comparison experiment in Euroc dataset
Experiment 4 | RMSE /cm |
---|
fr2_desk | fr2_xyz |
---|
fc_mono | 8.7 | 1.4 | SVO (with edgelets) | 9.7 | 1.1 | Semi-dense VO | 13.5 | 3.8 | Feature-based RGB-D SLAM | 9.5 | 2.6 | LSD-SLAM | 4.5 | 1.5 |
|
Table 4. Error comparison experiment in Tum dataset