Abstract: Visual measurement is widely used for the position and orientation measurement of roadheaders in coal mines due to its simple system and non-contact nature. However, the accuracy of visual measurement is affected by the calibration precision of camera parameters. The complex working conditions underground, uneven lighting, and hazardous areas lead to low calibration accuracy and difficulties in calibrating the cameras of the visual measurement system for roadheader position and orientation. To improve the automation level, calibration accuracy, and calibration speed of the camera calibration in underground roadheader pose visual measurement systems, an online automatic camera calibration method for roadheader pose vision measurement systems is proposed. Using the infrared target of the roadheader pose visual measurement system as the calibration target, a virtual point-line feature is constructed on the target plane and its corresponding image plane based on the constraints of the conic pole-polar line and the invariance of the cross ratio in projective transformation of straight lines. This allows the mapping of point-line features between the target plane and its corresponding image plane. A camera parameter homography solving model based on point-line dual feature constraints is then established to solve the linear solution for the camera parameters. Considering the influence of lens distortion, the minimisation of the re-projection error for point features and the line-to-line distance error for line features are introduced. A combined objective function is established, incorporating dynamic weight coefficients and constraints. Using the linear solution for the camera parameters derived from the homography solving model as the initial optimisation value, a nonlinear iterative optimisation algorithm is applied for further refinement, ultimately obtaining the optimised camera calibration results. A simulation analysis was conducted to examine the impact of the number of calibration views and noise on the calibration accuracy of different methods. Compared to other approaches, this method achieves convergence of calibration error with fewer views and demonstrates better noise robustness. An experimental platform for roadheader pose visual measurement was established, and camera calibration experiments were conducted using different methods. This method was applied to complete the pose measurement experiment. The results show that the camera parameters obtained using this method have a small error compared to the true values. The relative errors for f_x, f_y, u₀, v₀, k₁ and k₂ are 1.155%, 1.144%, 0.463%, 0.450%, 1.887%, and 4.082%, respectively. The average re-projection error is 0.197 pixel. The pose measurement experiment verified the effectiveness of the calibration method and successfully achieved online camera calibration for the underground roadheader pose visual measurement system, and provides an important support for high-precision pose visual measurement.