D$^3$FlowSLAM: Self-Supervised Dynamic SLAM with Flow Motion Decomposition and DINO Guidance

  • 2024-08-21 02:45:17
  • Xingyuan Yu, Weicai Ye, Xiyue Guo, Yuhang Ming, Jinyu Li, Hujun Bao, Zhaopeng Cui, Guofeng Zhang
  • 0

Abstract

In this paper, we introduce a self-supervised deep SLAM method that robustlyoperates in dynamic scenes while accurately identifying dynamic components. Ourmethod leverages a dual-flow representation for static flow and dynamic flow,facilitating effective scene decomposition in dynamic environments. We proposea dynamic update module based on this representation and develop a dense SLAMsystem that excels in dynamic scenarios. In addition, we design aself-supervised training scheme using DINO as a prior, enabling label-freetraining. Our method achieves superior accuracy compared to otherself-supervised methods. It also matches or even surpasses the performance ofexisting supervised methods in some cases. All code and data will be madepublicly available upon acceptance.

 

Quick Read (beta)

loading the full paper ...