深層学習に適した商品パッケージの認識精度の検証

抄録

The convenience store industry is facing a severe labor shortage due to the declining working population caused by the falling birthrate and aging population. Automation of display and disposal tasks is required to reduce employee burden. To achive the tasks, it is necessary to recognize products’ type, position and posture. Existing methods for product recognition include the use of depth sensors, Aruco markers, and deep learning. However, depth sensors are unable to identify objects with the same surface shape. And Aruco markers have problems such as spoiling the appearance of the product and making recognition difficult in dense conditions. And, deep learning takes a long time to learn a products. Therefore, a new package design for recognition using deep learning was proposed. This paper investigated its recognition accuracy and then reports the results.