DM-SLAM: A Visual SLAM Towards Dynamic Environment

【作者】 李奥； 王纪凯； 陈宗海；

【Author】 Li Ao;Wang Jikai;Chen Zonghai;Department of Automation,University of Science and Technology of China;

【机构】 Department of Automation,University of Science and Technology of China；

【摘要】 Simultaneous Localization and Mapping(SLAM) technology is proposed to solve the problem of robot self-positioning and environmental cognition.Over the past decades,many impressed SLAM systems have been developed and achieved good performance under certain circumstances.However,SLAM technology still have difficulty in handling all kinds of environments,such as extraordinarily dynamic or rough environments.In this paper,a semantic visual SLAM based on ORB-SLAM2 towards dynamic environments named DynaSLAM2 is proposed.Our proposal is to use Mask R-CNN to pixel-wise segment the priori dynamic objects in the frames(e.g.,walking people),so that the SLAM algorithm does not have to extract features on them.DynaSLAM2 outperforms the accuracy of standard visual SLAM baselines and a high robustness in highly dynamic scenarios.And it also estimates a map of the static parts of the scene,which is a must for long-term applications in real-world environments.更多还原

【Abstract】 Simultaneous Localization and Mapping(SLAM) technology is proposed to solve the problem of robot self-positioning and environmental cognition.Over the past decades,many impressed SLAM systems have been developed and achieved good performance under certain circumstances.However,SLAM technology still have difficulty in handling all kinds of environments,such as extraordinarily dynamic or rough environments.In this paper,a semantic visual SLAM based on ORB-SLAM2 towards dynamic environments named DynaSLAM2 is proposed.Our proposal is to use Mask R-CNN to pixel-wise segment the priori dynamic objects in the frames(e.g.,walking people),so that the SLAM algorithm does not have to extract features on them.DynaSLAM2 outperforms the accuracy of standard visual SLAM baselines and a high robustness in highly dynamic scenarios.And it also estimates a map of the static parts of the scene,which is a must for long-term applications in real-world environments.更多还原