MD-SLAM: A Multi-Task Deep-Learning-Based Visual SLAM System in Dynamic Environments

With the development of self-driving vehicles and intelligent robots, visual simultaneous localization and mapping (SLAM) has attracted significant attention. In dynamic environments, non-stationary objects can cause performance degradation of visual SLAM systems. The inaccurate pose and lack of semanticinformation in feature points can lead to incorrect differentiation between static and dynamic points, resulting in a degradation of the visual SLAM system's performance. In this paper, we propose a novel visual SLAM system based on multi-task deep neural networks to address this issue. Specifically, we apply multi-task deep neural networks to extract higher quality and more robust oriented local features and perceive dynamic semantic regions, which are used to better remove dynamic points spatially and potential dynamic points semantically in the SLAM system. We evaluate our method on public datasets, and the results show that our method outperforms existing visual SLAM systems.