A two-camera vision system with far and near sight distance for global target positioning and accurate local measurement is usually required when an industrial robotic arm works in multiple stations. Lacking a common field of view, the intrinsic and extrinsic parameters of the vision system cannot be calibrated using traditional methods; therefore, a new calibration technology based on point-array coded targets and multi-view geometry is proposed. First, two sizes of calibration panels with point-array coded targets matching the field of view size are prepared for the two-camera vision system with far and near sight distance. A sub-pixel edge detection method based on gray gradient is proposed to improve the positioning accuracy of the ellipse centers. Then, a decoding algorithm of the point-array coded target is improved for robust image feature matching. Based on this, using high-precision central pixel coordinates of point-array coded targets under different perspectives and the corresponding relationship of targets between different images, the intrinsic parameters of two cameras and the transformation matrices between cameras and calibration panels can be calculated using multi-view geometry technology. Then, a hand-eye calibration equation, AX=ZB, is constructed from the transformation matrix, and the extrinsic parameter X between two cameras can be obtained by solving the equation. Finally, the suppression of calibration error by two kinds of optimization methods is analyzed. Experimental results demonstrate that the calibration accuracy is improved after optimization. The angle error is reduced to 0.05°, and the position error is reduced to 0.36 mm.