Robust Visual SLAM Algorithm for Dynamic Indoor Environments

In real scenes, traditional visual SLAM algorithm is limited by the assumption of static environment. Due to the influence of moving objects, the traditional visual odometry makes a large number of mismatches and makes the system unable to run statically in real scenes. In this paper, a robust visual SLAM algorithm for indoor dynamic environment is proposed based on deep learning. Firstly, the object detection network is used to detect the dynamic objects and determine the moving objects in the surrounding environment. Secondly, a semantic data association method and key-frame selection strategy are proposed for the tracking thread and the local graph thread to reduce the influence of moving objects on the algorithm accuracy. Finally, the TUM open datasets is validated. Experimental results show that the proposed method reduces the mean root mean square error (RMSE) of ORB-SLAM2 by more than 94% in dynamic scenarios compared to the previous improved ORB-SLAM2. In addition, compared to other deep learning based SLAM systems, the proposed method has a better balance of speed and accuracy in dynamic environments.