Enhancing SLAM Accuracy in Urban Dynamics: A Novel Approach with DynaVINS on Real-World Dataset

Simultaneous Localization and Mapping (SLAM) is a crucial algorithm in autonomous robot navigation, enabling systems to simultaneously map the environment in real-time while estimating their position and orientation. One notable implementation of SLAM is the Visual-Inertial Navigation System (VINS), which enhances trajectory estimation accuracy by combining visual data from cameras with inertial data from IMUs. However, traditional VINS approaches often struggle in dynamic environments where moving objects can degrade performance. To address this, DynaVINS, a visual-inertial SLAM method, is designed to robustly operate in such conditions by effectively separating static and dynamic features. This study evaluates DynaVINS using the KITTI and primary datasets, with primary metrics including Absolute Trajectory Error (ATE) and Relative Pose Error (RPE). The results demonstrate that DynaVINS effectively tracks relative pose changes (low RPE), but it faces challenges in achieving global accuracy, as indicated by significant ATE values, particularly in scenarios with sharp turns or abrupt orientation changes. Consequently, further optimization is required to improve global accuracy and ensure reliable performance in real-world scenarios, particularly in urban road environments.