Optical Character Recognition for Autonomous Stores
Pétia Georgieva, Pei Zhang
Abstract
Autonomous stores are a new trend in the retail world. Vision-based product identification is one of the frequently used sensing modalities. There are different approaches and design configurations for the vision sensing but they all are based on extracting features for the product appearance, such as color, shape, dimension. None of the proposed solutions rely on reading the package label of the product being identified. In this paper, we present an Optical Character Recognition (OCR) framework, inspired by natural language processing, for package label detection and transcription from product images. To the best of the authors knowledge, this is the first autonomous store system that considers an OCR sensing modality. This is a step forward to a closer approximation to the human cognitive way of purchasing products.