Objectives Current monocular visual inertial odometry (VIO) algorithms impose stringent requirements on device motion states and visual observation conditions during initialization, which are often challenging to meet in real-world engineering applications. A fast and stable initialization method is thus essential.

Methods The proposed method utilizes the known edge length of ArUco markers to recover depth information through edge length constraints, thereby estimating the spatial relationship between the camera and the ArUco markers. Concurrently, the inertial measurement unit (IMU) is initialized under stationary conditions. By incorporating the pre-calibrated extrinsic transformation between the camera and IMU, the spatial relationship between the global coordinate system and the ArUco markers is established.

Results The accuracy of depth recovery at corner points was validated using profiles of the differences between the measured distance of two points and the recovered depth. Multiple experiments under both well-lit and low-light conditions showed that the mean error of this difference was 0.75 mm. The VIO positioning accuracy of the device within a certain spatial range after initialization was further evaluated, and several experiments demonstrated a positioning error of 0.009 m and an attitude error of 0.4°. Additional experiments were conducted to assess the stability and efficiency of the proposed method. Compared with the monocular VIO initialization methods used in VINS-Fusion, OpenVINS, and ORB-SLAM3, the proposed method achieved higher stability and efficiency, while also yielding a better overall trajectory in the subsequent VIO process.

Conclusions The proposed method demonstrates high efficiency and robustness in initialization, with notable improvements in positioning accuracy during subsequent VIO tracking.