针对单一传感器的局限以及视觉惯性融合算法计算复杂度等问题，提出一种基于深度估计半直接视觉里程计（SVO）与惯性测量单元（IMU）融合的定位算法。将深度估计模块集成至 SVO 中，采用扩展卡尔曼滤波器构建松耦合融合框架，融合视觉位姿信息与 IMU 加速度、角速度数据，实现状态估计与位姿校正。在 KITTI 数据集以及真实环境中验证表明，该算法的绝对轨迹误差低于纯视觉算法，且在复杂场景下预测轨迹与真实轨迹高度贴合。

To address the limitations of single sensors and the computational complexity of visual-inertial fusion algorithms， a positioning algorithm fusing semi-direct visual odometry （SVO） based on depth estimation with an inertial measurement unit （IMU） is proposed. A depth estimation module is integrated into the SVO and an extended Kalman filter is employed to construct a loosely coupled fusion framework. Visual pose information was fused with the IMU acceleration and angular velocity data in this framework to achieve state estimation and pose correction. Validation of the KITTI dataset and real-world environments demonstrated that the absolute trajectory error of the proposed algorithm was significantly lower than that of pure visual algorithms， and the predicted trajectory was highly consistent with the real trajectory in complex scenarios.

TN911.73；TP182

新重庆青年创新人才项目（CSTB2024NSCQ-QCXMX0096）