In environments where GPS signals are unavailable, SLAM algorithms relying solely on visual-inertial odometry can achieve local accurate positioning, but they suffer from significant accumulated errors during long-distance movements, leading to decreased positioning accuracy. Although GPS can provide global location information, its performance is often unstable in complex environments such as urban canyons, tunnels, and indoor spaces, where signals are easily blocked or interfered with, limiting its applicability. To address aforementioned issues, the VIG-SLAM algorithm is proposed, which integrates a tightly-coupled visual/inertial/odometer positioning system with GPS data. First, a GPS accuracy factor model and anomaly detection mechanism are developed to evaluate and dynamically select high-quality GPS data suitable for fusion. Second, an improved adaptive time difference compensation strategy is proposed to solve the problem of timestamp mismatch between GPS and VIW systems. At the same time, the weight of GPS signal is dynamically adjusted in time difference compensation to improve positioning accuracy and robustness in complex environments. Finally, a global pose graph optimization model with GPS constraints is constructed, using GPS global positioning information as a global constraint to complement VIW local positioning, achieving robust positioning in large-scale environments. The proposed method’s effectiveness is validated on public datasets and real-world experimental scenarios, with results showing that the average positioning accuracy of VIG-SLAM algorithm improves by at least 15% compared to current mainstreamvisual SLAM algorithms, demonstrating strong robustness and accuracy advantages.