Continuous localization method for unmanned mobile platform based on GNSS/IMU/Vision fusion in multi-stage environments of coal mine

Accurate localization of unmanned platforms in complex and dynamic mining environments is increasingly essential amid the rapid advancement of intelligent mining. Traditional GNSS/IMU localization methods face several limitations in mining applications, including low adaptability to dynamic environments and poor performance in GNSS-denied areas such as indoor storage zones. To overcome these challenges, this study proposes a GNSS/IMU/Vision fusion framework for localization of unmanned mining mobile platforms. To handle intermittent GNSS signals during multi-stage transitions, a Vision–IMU integrated localization method is developed based on point–line feature extraction. Image enhancement techniques are employed to address issues of low illumination and motion blur. A point–line feature fusion strategy is designed, incorporating line constraints in weak-texture areas to improve robustness. In GNSS-available environments, a factor-graph-based GNSS/IMU/Vision fusion localization method is constructed. By flexibly configuring the factor graph, the framework enhances fault tolerance and ensures consistent global optimization. The proposed method is experimentally validated in open areas, simulated tunnel environments with GNSS denial, and hybrid scenes with varying GNSS signal strength. Experimental results show that the proposed system achieves a root mean square error (RMSE) of 0.3483 m, outperforming the conventional Visual-Inertial Odometry (VIO) and Global Visual-Inertial Navigation System (GVINS) methods. The framework demonstrates high-precision and robust localization performance for unmanned mining platforms under multi-stage GNSS signal variations.