To improve the automation level and measurement accuracy of the visual measurement system, a novel method based on template matching and peaks of gradient histogram is proposed to extract the corners of diagonal markers automatically. The rotation-invariant template matching method is used to get the coefficient correlation matrix of the original image and the standard template, and the candidate positions of markers are obtained by two thresholding selection. The false candidates are eliminated and the initial coordinates of markers are found depending on the properties, which are two lines intersecting at the center of the marker and their gray gradient of the marker with multiple peaks. The subpixel is located by generating ideal templates of diagonal markers and applying the correlation coefficient fitting extreme value method. The experimental results show that the proposed method can automatically extract the corners of diagonal markers or checkboard images in a complex environment.This method has the merits of few adjustable parameters, strong robustness, precision and universality. It can be used in measuring situations where the environment light source greatly in engineering practice.