Research on RGB-D Visual SLAM Algorithm Based on Adaptive Target Detection

Most of the traditional SLAM systems are based on static environment assumptions, but they are easily affected by dynamic targets in real environments, resulting in a serious degradation of the robustness and accuracy of the algorithms. In this paper, we focus on visual SLAM systems in dynamic scenes, introducing an object detection network in SLAM to obtain the low-level semantic information of dynamic targets, and adopting a new dynamic point selection strategy to classify the detected targets into three motion types, and then fusing the semantic information to eliminate the dynamic feature points. Experiments show that the proposed method outperforms traditional methods in dynamic scenarios, and the real-time performance of the proposed method is improved compared with the semantic segmentation-based SLAM system.