부산대학교 도서관

Visual–inertial systems suffer from the issue of error accumulation, which limits their application to small-scale and short-duration scenarios. To enhance the applicability of visual–inertial odometry (VIO), we propose T-VIO, a filter-based estimator that tightly integrates visual, inertial, and time-differenced carrier phase (TDCP) measurements for low-drift and high-precision local pose estimation. The introduction of TDCP suppresses the error accumulation and provides absolute heading observability. On this basis, the relative pose and absolute attitude provided by T-VIO are fused with global navigation satellite system (GNSS) solutions through an optimization-based framework to achieve global consistent pose estimation. Real-world experiments were conducted in both campus and urban environments to evaluate the performance of the proposed method. The results show that our proposed method effectively constrains time-increasing navigation error in VIO and demonstrates its applicability in large-scale scenarios with complex conditions. In addition, the experiments indicate that even with intermittent global positions, our system can still achieve consistent and accurate global pose estimation.