Abstract: In order to online monitor the physical stability of the imaging environment for cigarette packet appearance inspection devices, an in-situ measurement method based on computer vision was proposed, taking the advantage of the fact that most of these devices are equipped with three cameras. 1)The LK luminous flux method was used to analyze the continuous images collected in real time by the front camera to obtain the vibration amplitude data of the device. 2)The two side cameras were used to form a binocular vision system, the posture change data of the cigarette packets were obtained by combining with camera calibration and algorithms such as Hough transform. 3)The average vibration amplitude μ_y, the average rotation angle μg, and the average offset distance μ_d were created as criteria for discriminating the status of the device when the vibration amplitude y_i > 1.5 μ_y, the vibration of the device is defined as an abnormal vibration; if the rotation angle g_i > 1.5 μ_g or the offset distance d_i > 1.5 μ_d, the posture of the cigarette packet is defined as an abnormal posture. The method was verified and its performance was tested on 16 devices. The results showed that this method could accurately detect the abnormal vibration of the device and the abnormal posture of the cigarette packet in real time. The correlation coefficient between the results measured by this method and an accelerometer was higher than 85%, and the compound measurement error for the posture changes(including rotation and translation)was less than 2%. The ability of different devices to control the posture of the cigarette packet varied widely, and the vibration amplitude of the devices increased as the increase of their length of service. This method provides technical support for improving the appearance inspection of cigarette packets.