关键词: 同时定位与地图创建（SLAM）, 室内动态环境, ORB-SLAM2, 动态检测

Abstract:

As a basic function of autonomous mobile robots, Simultaneous Localization and Mapping（SLAM） has been widely researched in recent years. However, most state-of-art visual SLAMs adopt a strong scene rigidity assumption for analytical convenience, which limits the utility of these algorithms for real-world environments with independent dynamic objects. This paper presents a robust visual SLAM towards dynamic indoor scenes, which is built on the RGB-D mode of ORB-SLAM2. A dynamic detection method based on geometric constraints is added to the front end of ORB-SLAM2. First, the dynamic features in the scene are coarsely filtered using a geometric constraint method. Then the remaining features are used as sample points for the improved Random Sample Consensus（RANSAC） algorithm to estimate the stable fundamental matrix. And the epipolar geometry is used to filter out the real dynamic features in the scene. Experiments on the public TUM RGB-D dataset are conducted to evaluate the proposed approach. This evaluation reveals that the proposed algorithm can effectively improve the positioning accuracy of the ORB-SLAM2 system in high-dynamic scenarios.

Key words: Simultaneous Localization and Mapping（SLAM）, dynamic indoor environment, ORB-SLAM2, dynamic detection