Instance Segmentationを利用した屋内の動的環境における潜在的動物体に基づくSLAMの改善

抄録

In this paper, we improve a robust SLAM for dynamic environments by using object detection considering potential dynamic objects. Generally, dynamic environments have a negative impact on SLAM because correspondence between frames is wrong and constructed map includes ghosts. Conventional methods resolved this problem by predefining dynamic objects and removing them, however these cannot handle degenerate environments. Therefore, we define potential dynamic objects to increase the number of features as much as possible and to improve the problem of degeneracy that occurs in conventional methods. In this study, we use instance segmentation, which can detect objects more accurately at the pixel level, and verify its effectiveness.