In many practical application scenarios, the camera calibration is not highly accurate due to unclear calibration images and low target detection accuracy, which limits the improvement of measurement accuracy. To solve this problem, a binocular camera calibration method based on sub-pixel edge detection is proposed. The initial integer-pixel edge values of the target identification points are obtained by an adaptive double-threshold Canny operator. In addition, the initial integer-pixel edge values are taken as the center to estimate the second-order edge parameters of a discontinuous edge model based on the partial area effects, and ellipse fitting on the set of sub-pixel edge points is performed to obtain the accurate position of the target identification points. Finally, the set of the identification points used to solve the calibration parameters is obtained by correcting the sorting position of these points, which thus achieves the high precision calibration of cameras under complex environments. The test experiments on typical calibration scenarios show that, compared with the existing methods, the proposed method can improve the calibration accuracy by 23% in a conventional environment and 68% in a high-temperature oven environment with low contrast and relative resolution, respectively.