A CNN-Based Identification Method for Products Appearing in Panoramic Images

摘要

A mixed-reality web shopping system using in-store panoramic images can provide customers with a highly realistic shopping experience on a web browser. To implement this system, a product chosen by a customer from a panoramic image must be identified automatically. Although we previously proposed a product identification system using a convolutional neural network (CNN), it does not work well enough. The study in the current paper proposes a novel method to identify products by using multiple images in panoramic images, in which the product appears multiple times. From experimental results, it is verified that the performance of the system is improved by using this method.