Binocular Vision Localization Method for Roadheader Based on Point-Line Features in Coal Mines

Accurate and robust localization of roadheaders is essential for intelligent control in underground coal mining, where visual perception systems face significant challenges such as low illumination, abrupt lighting changes, dust interference, and complex backgrounds. This paper proposes a stereo vision-based localization method for roadheaders that integrates point and line features to enhance pose estimation accuracy and robustness under complex lighting conditions. To address image degradation caused by the coexistence of low-light and overexposed regions, an adaptive multi-region image enhancement approach is developed. The method dynamically segments the image into high-, medium-, and low-illumination regions based on average brightness and standard deviation, applying region-specific enhancement strategies such as gamma correction, CLAHE, and contrast adjustment. A gradient-based weighted fusion mechanism and bilateral filtering are introduced to improve image quality and feature extraction stability. Line features are extracted using EDLines and represented via Plücker coordinates. A joint point-line optimization model with a dynamic weighting scheme is constructed to improve localization robustness. Experiments on public datasets including EuRoc MAV and KITTI demonstrate that the proposed method achieves an APE of 0.10 m and RMSE of 0.13 m, outperforming existing methods such as PL-SLAM and PL-EDLines-SLAM. Further validation in a dust- and glare-simulated underground environment yields an APE of 0.15 m and RMSE of 0.17 m, confirming the method’s environmental adaptability and practical applicability. This study provides a reliable visual localization solution for underground roadheaders, supporting downstream tasks such as path planning, tracking control, and autonomous operation.