Author Affiliations
School of Instrument Science and Engineering, Southeast University, Nanjing 210096, Jiangsu, Chinashow less
Fig. 1. General framework of proposed algorithm
Fig. 2. Line feature extraction results and time consumption of each algorithm in measured environment. (a) Hough algorithm (118.045 ms); (b) LSD algorithm (62.7 ms); (c) LSWMS algorithm (40 ms); (d) EDLine algorithm (18.2 ms)
Fig. 3. Possible changes in same line feature between two consecutive frames. (a) Existence of slight angles; (b) existence of slight distances
Fig. 4. Schematic diagram of coordinate frame involved in proposed system
Fig. 5. Platform for actual measurement data acquisition
Fig. 6. Line feature matching results of traditional LSD algorithm and proposed algorithm. (a) LSD algorithm; (b) proposed algorithm
Fig. 7. Trajectory fitting curves of KITTI data set experiments. (a) 09_30_0018 data set; (b) 09_30_0027 data set
Fig. 8. APE root mean square error (APE_RMSE) comparison curves of KITTI data set experiments. (a) 09_30_0018 data set;
Fig. 9. APE_RMSE comparison curves of tunnel data set experiments. (a) Tunnel 1 data set; (b) Tunnel 2 data set
Fig. 10. Trajectory fitting curves of urban road data set experiment. (a) Urban road data set; (b) road of test field data set
Fig. 11. APE_RMSE comparison curves of urban road data set experiment. (a) Urban road data set; (b) road of test field data set
Sequence | Vins_Mono | PL-VIO | Vins_Fusion | Proposed |
---|
09_30_0018 | 20.46 | 11.18 | 5.37 | 1.03 | 09_30_0027 | 26.41 | 23.32 | 3.66 | 0.74 | 09_30_0033 | 22.86 | 20.39 | 7.25 | 2.54 | 09_30_0034 | 15.31 | 26.78 | 4.62 | 0.68 |
|
Table 1. Comparison of APE_RMSE of each algorithm in KITTI data set
Sequence | Vins_Mono | PL-VIO | Vins_Fusion | Proposed |
---|
Tunnel 1 | 86.24 | 65.62 | 73.66 | 60.94 | Tunnel 2 | 70.19 | 46.26 | 64.62 | 34.44 | Tunnel 3 | 165.53 | 96.06 | 144.32 | 70.56 |
|
Table 2. Comparison of APE_RMSE of each algorithm in tunnel data set
| Urban road | Road of test field |
---|
| APE_RMSE /m | Completeness /% | APE_RMSE /m | Completeness /% | SPP | - | 92.86 | - | 84.00 | Vins_Mono | 206.73 | 100.00 | 98.19 | 100.00 | PL-VIO | 151.42 | 100.00 | 82.58 | 100.00 | Vins_Fusion | 99.90 | 100.00 | 63.09 | 100.00 | Proposed | 3.10 | 100.00 | 14.72 | 100.00 |
|
Table 3. Positioning performance comparison of APE_RMSE of each algorithm in urban road data set
Operation | PL-VIO /ms | Proposed /ms |
---|
Point feature extraction and tracking | 7.860 | 6.011 | Line feature extraction and tracking | 31.376 | 27.219 | Line feature elimination | 0.027 | 0.009 | Marginalization | 1.326 | 1.192 |
|
Table 4. Comparison of real-time performance of proposed algorithm with PL-VIO algorithm