Size measurement method of cigarette pack based on key point detection and stereo matching

In modern production, the process of gluing and forming the outer package of cigarette packs has been fully automated. However, due to the influence of equipment operation deviation, the molded cigarette packs may be shifted during the gluing process, causing the final outer packaging to fail quality standards. Currently, cigarette factories mainly rely on manual measurement to inspect critical dimensions of the cigarette. To reduce human resource consumption and enhance inspection accuracy, a method for measuring the part dimensions of the cigarette pack based on key point detection and stereo matching is proposed. Specifically, the YOLOv8 algorithm with excellent real-time performance is first used to recognize the Lotus cigarette pack in the picture. The Lite-HRNet algorithm, which has the ability of high-resolution feature extraction, is used to detect the key points of the recognized cigarette pack, thus obtaining the 2Dpixel coordinate information of the parts to be detected. Next, utilizing the depth information provided by the CREStereo algorithm, the 2D pixel coordinates of the keypoints are converted to 3D camera coordinates. Finally, the spatial distances between key points are calculated based on the Euclidean distance formula to produce accurate measurements. In the actual measurement of 20 packs of cigarettes, the average absolute errors of the distances between the left and right side silver waistlines were 2.68 mm and 2.34 mm, respectively. This paper presents a method that can effectively meet the practical needs for measuring the dimensions of cigarette pack parts.