Real-time visual SLAM based on semantic information and geometric information in dynamic environment

Abstract Simultaneous Localization and Mapping is the core technology for mobile robots to autonomously explore and perceive the environment. However, dynamic objects in the scene significantly impact the accuracy and robustness of the visual SLAM system, limiting its applicability in real-world scenarios. To address this problem, this paper proposes a real-time RGB-D visual SLAM algorithm designed for indoor dynamic scenes. In this paper, a parallel lightweight object detection thread is designed to detect potential moving objects and generate 2D semantic information based on the YOLOv7-tiny network. Then, a new dynamic feature removal strategy is proposed in the tracking thread, which tightly integrates semantic information, geometric constraints, and feature point depth-based RANSAC to effectively reduce the impact of dynamic feature. To evaluate the effectiveness of the algorithms discussed in this paper, we conducted comparative experiments using other state-of-the-art algorithms on the TUM RGB-D dataset and Bonn RGB-D dataset, as well as in real-world dynamic scenes. The results demonstrate that the algorithm maintains excellent accuracy and robustness in dynamic environments, while also exhibiting impressive real-time performance.