Common Contracts
1 similar Grant Agreement contracts
Grant agreement No: 101017008Grant Agreement • October 5th, 2022
To decide which semantic classes would benefit localization, we estimated how likely they are to move around. We prepared a semantic map for all detectable classes, and examined the training-dedicated recordings T1-T5 spanning over multiple weeks. As our dataset includes the ground truth pose of the robot using an external reference system, we were able to con- clude whether the position of detected objects corresponded to their position in the map. Using Eq. (4), we consider a detected object to correspond to the semantic map if d < τs. We calculate the ratio of map-consistent detection per semantic class, and deem a semantic class stable if the ratio was above
