Multi-Sensor Information Fusion Methods for Coal Mine Exploration in GNSS-Denied Scenarios

In coal exploration scenarios such as underground mines and enclosed tunnels, Global Navigation Satellite System (GNSS) signals are often unavailable due to geological obstacles, while challenges like dust interference, dynamic obstacles, and extreme lighting variations further degrade localization accuracy. To address these issues, this paper proposes a novel multi-sensor fusion framework named LiVIC-EKF (LiDAR-Visual-Inertial Covariance-adaptive Extended Kalman Filter). The algorithm integrates LiDAR, binocular cameras, and an Inertial Measurement Unit (IMU) to achieve robust six-degree-of-freedom pose estimation. Key steps include offline calibration of sensor parameters, extraction of LiDAR geometric features using an improved LOAM algorithm, and tight-coupling optimization of visual-inertial odometry via VINS-Fusion. The LiVIC-EKF framework adaptively weights the contributions of LiDAR and visual-inertial odometry through a dynamic covariance mechanism, ensuring stable localization in complex environments. Experimental results on the KITTI dataset demonstrate that LiVIC-EKF reduces the mean Absolute Pose Error (APE) to 2.60 meters and standard deviation to 1.66 meters, outperforming state-of-the-art methods such as A-LOAM, FAST-LIO2, and VINS-Fusion. This approach provides a reliable solution for coal exploration applications where GNSS is unavailable.