The positioning accuracy of simultaneous localization and mapping (SLAM) system based on visual and IMU fusion depends on accurate IMU initialization and external parameters between camera and IMU to a large extent. Hence, the SLAM system recalibrated when the relative position of the camera and IMU changed. Unfortunately, the off-line calibration of the SLAM system is complicated work. To solve the problem, a new method of external parameter self-calibration and IMU online initialization devised. Moreover, the proposed method employs the calibrated stereo camera to avoid the ambiguous-scale of the monocular camera. Four processes are included in our method. Firstly, the parameter of external rotation between camera and IMU with the rotation between camera and IMU of two consecutive frames as input is calculated. Secondly, the gyro bias estimated by the obtained external parameter. Thirdly, the gravity and the parameter of external translation roughly estimated without considering the accelerometer bias. Finally, the gravity and external translation optimized with the accelerometer bias. Experiential results on datasets demonstrate that the proposed method with stereo VI-SLAM has better performance of positioning accuracy, convergence, and deviation of external parameters than that of the state-of-the-art VINS-mono of the new version.
展开▼
