Abstract: Given the problem of the low computational efficiency and accuracy of pose estimation in the visual-inertial simultaneous localization and mapping (VI-SLAM) system, a tightly-coupled stereo VI-SLAM algorithm combining an extended Kalman filter and incremental bundle adjustment is proposed. In the front end, an extended Kalman filter is employed to couple the measurement information of the inertial measurement unit and the stereo camera to estimate the pose and velocity. In the back end, the pose is optimized by incremental bundle adjustment to obtain a globally consistent motion trajectory. In comparison with the SLAM algorithm that only uses the filtering method or the optimization method, the proposed algorithm can improve the accuracy of pose estimation by reducing the linearization error of pose estimation and the occupation of computing resources on the public dataset EuRoC. A mobile robot is used to test the algorithm in a real environment, and our results verify the practical feasibility of the algorithm.