Improved visual SLAM algorithm in dynamic environments incorporating optical flow and semantic information

We propose a SLAM (Simultaneous Localization and Mapping) method for dynamic environments based on the fusion of semantic and optical flow information. Traditional visual SLAM systems suffer a decline in localization accuracy due to interference from moving objects in the scene. To address this issue, we employ a method that combines semantic and optical flow information. First, we extract semantic and optical flow information from the scene using the deep learning network. Then we divide the scene into different regions based on the semantic information. Finally, we employ suitable fusion methods according to the different regions to identify and eliminate the interference of moving objects on localization. We validated and assessed the performance of the proposed method on the public TUM dataset. The final experimental results demonstrate that the proposed method exhibits excellent performance in dynamic scenarios. This is of significant importance for improving the performance of the SLAM in dynamic environments.