The work aims to rapidly realize the machine vision measurement of the size of packaging boxes. Based on machine vision technology, according to the imaging principle of binocular camera, the SUSAN corner extraction algorithm combined with checkerboard calibration method was used to obtain the internal and external parameters of camera and image distortion correction. The two images on the left and right were matched by the SIFT matching operator. The corresponding matched feature points were found. The parallax values of matching point pairs were used to restore the image depth information, thus calculating the length, width and height of the packaging box. The result showed that the packaging boxes of different postures could be rapidly measured. Compared with the physical value, the error of machine vision detection value was about 1 cm, meeting the detection requirements. Featured by high matching accuracy and strong robustness, etc., the proposed method can rapidly measure the size of packaging boxes of different postures.