Research Article
An End-to-End Robotic Visual Localization Algorithm Based on Deep Learning
Table 2
Comparison of localization errors on the Cambridge Landmarks dataset.
| | Scene | Area or volume | PoseNet | Bayesian PoseNet | Ours (tra/m,ori/°) |
| | King’s College | 5,600 m2 | 1.92 m, 5.40° | 1.74 m, 4.06° | 1.33 m, 1.73° | | Old hospital | 2,000 m2 | 2.31 m, 5.38° | 2.57 m, 5.14° | 1.46 m, 4.23° | | Shop facade | 875 m2 | 1.46 m, 8.08° | 1.25 m, 7.54° | 1.14 m, 6.33° | | St Mary’s Church | 4,800 m2 | 2.65 m, 8.48° | 2.11 m, 8.38° | 1.42 m, 5.68° | | Average all | – | 2.08 m, 6.83° | 1.92 m, 6.28° | 1.34 m, 4.49° |
|
|
