Multi-Modal Visual Place Recognition in Dynamics-Invariant Perception Space

Visual place recognition is one of the essential and challenging problems in fields robotics. In this letter, we for first time explore use multi-modal fusion semantic visual modalities dynamics-invariant space to improve dynamic environments. We achieve by designing a novel deep learning architecture generate static segmentation recover image directly from corresponding image. then innovatively leverage spatial-pyramid-matching model encode into feature vectors. parallel, encoded using popular Bag-of-words model. On basis above features, finally measure similarity between query target landmark joint their codes. Extensive experiments demonstrate effectiveness robustness proposed approach