DGO-VINS: A Visual-Inertial SLAM for Dynamic Environments With Geometric Constraint and Adaptive State Optimization

Traditional SLAM performs well in static environments, but experiences degeneration of localization accuracy and stability in dynamic settings. To enhance performance in dynamic environments, this letter presents DGO-VINS, a real-time dynamic visual-inertial SLAM system based on geometric constraint and moving probability. First, we propose a semantic-integrated feature detection method that adaptively adjusts the feature detection area, effectively addressing the scarcity of static features in dynamic environments. Subsequently, we introduce a dynamic feature rejection algorithm that integrates Inertial Measurement Unit (IMU) measurements, semantic information, and multi-view geometric constraint, accurately and efficiently removing dynamic features. Finally, to incorporate potential dynamic features into the optimization process robustly, an adaptive state optimization method is proposed. This method uses a binary Bayesian filter to estimate the moving probability of potential dynamic features and constructs an adaptive loss function based on this moving probability. The proposed system was evaluated on multiple publicly available datasets, as well as in real-world environments, to verify its effectiveness. The experimental results demonstrate that the proposed method surpasses existing approaches, delivering improved localization accuracy and greater robustness in a variety of dynamic environments.