Abstract: In the vision measurement system for cutting head position of cantilever roadheader based on infrared LED characteristics, the stability of external calibration and the accuracy of infrared LED spot center extraction have an important influence on the cutting head position detection accuracy. The existing external parameter calibration method relies on experience to swing the cutting arm to the center position (unknown), and the calibration results have large fluctuations. In order to solve the above problem, a multi-point fixed external parameter calibration method is proposed. This method controls the cutting arm of the roadheader to swing to the four known limit positions of upper left corner, upper right corner, lower left corner and lower right corner respectively, and collects the target images. The method calculates the value of the external parameter matrix, which can improve the stability of the external parameter calibration effectively. The existing gray-scale centroid method uses the grayscale value of the pixel as the weight to calculate the spot centroid. And the accuracy can only reach the pixel level, which only roughly meets the practical application requirements. In order to solve this problem, a sub-pixel edge detection algorithm is proposed to improve the spot center extraction method. Firstly, the gray-scale centroid method is used for coarse extraction of the spot center. Secondly, the sub-pixel level edge detection algorithm is used to find the sub-pixel level edge coordinates. Finally, the least squares method is used to fit the spot center to achieve accurate extraction of the spot center. The experimental results show that the improved spot center extraction method reduces the maximum measurement error of the target LED lamp spacing from 3.2 mm to 1 mm, which improves the detection accuracy. The external parameter matrix obtained by the multi-point fixed external parameter calibration method is relatively stable, the maximum variation of displacement in the translation matrix is 15 mm, and the maximum variation of angle in the rotation matrix is 1°. Before the improvement of the vision measurement system, the measurement error of the cutting head swing angle was within ［-1.2°,1.7°］. After the improvement, the error of the horizontal swing angle of the cutting head is within ［-0.5°,0.5°］ and the error of the vertical swing angle is within ［-0.6°,0.6°］. The results show that the improved method improves the detection accuracy of the cutting head swing angle effectively.