CNN-based shoe-upper pattern recognition and generation of adhesive point

Abstract

Although the manpower participating in the shoe-upper sewing process is more than 70% of the entire process, there is a shortage of manpower due to the lack of worker skills and the disadvantages of many avoidance processes. Globally, many companies have proposed a way to automate fuse sewing technology to solve this problem. It is necessary to fix the shoe-upper patterns before bonding the shoe-upper in the automated shoe-upper manufacturing process. We propose a method to create a path point for the adhesive application of the vision-based shoe-upper. We conducted experiments using vector images and real images. In the proposed method, the convex hull algorithm is applied to the input image to detect the shoe-upper region. Next, two center points are set in the shoe-upper region and a center line connecting the center points is detected. Then, a grid pattern is formed at regular intervals based on the direction of the center line, and the intersection of the grid pattern is detected. Finally, we removed the outlier of the intersection point of the grid pattern and the intersection points, except for the outlier, were set as the path points of the adhesive and recognized the shoe-upper pattern with 85% accuracy using a convolutional neural network (CNN). © ICROS 2017.