Lens Assembly

This material shows the actual assembly process of video lens production and was filmed in a simulated field. The research data consisted of nine operators performing the process of assembling video footage. Each video contained a complete SOP action and was repeated several times.The data includes a total of 90 videos, including 72 videos with actions within standard operating procedures and 18 videos with actions that are not within standard operating procedures, totaling approximately 270,000 frames, with a total length of approximately three hours, and an average video length of approximately one hundred seconds. To collect video data, this study used a Microsoft LifeCam Studio camera to shoot from a top-down perspective with the operator as the center, ensuring that the camera can clearly capture the operator's upper body and hands, such as assembling a dataset produced by shooting video footage at a speed of 30 frames per second and a resolution of 1280\u00d7720.In this dataset, each SOP contains thirteen steps, divided into six different actions, corresponding to six labels, namely Waiting, Brush, Assemble, Screw, Scan, and Sticker. The product appearance is based on the components of an actual video lens, and includes the following items: lens body (gripped by a robotic arm), top cover, right side cover, left side cover and sticker. First, the operator applies adhesive to the surface of the lens, then assembles the cover on the lens and tightens the screws, repeating this action three times, and finally scans and sticks the sticker. In each SOP, the execution method and execution time of each action are different according to the habits of different operators, but the order and processing position are kept the same. Common non-standard operation procedure actions in industry include Texting, Taking a Notes and Applying Tape, etc.After the video data is captured, domain experts who have industrial management knowledge and are familiar with this SOP will annotate the video. Each frame in a movie has its corresponding label, so each movie has multiple action labels. This annotation method helps to accurately analyze and evaluate the execution of various actions and the recognition ability of the model. It is worth noting that the same label is regarded as the same action, regardless of the assembled objects. For example, assembling the top cover and assembling the side cover are both regarded as assembly. This method helps to simplify the action classification process and improve the model's recognition accuracy for different actions. At the same time, it avoids labeling errors caused by subtle differences and ensures data consistency and reliability.