See-Shop System: An Assistive Outdoor Navigation System for the Visual Impaired Based on Deep Learning Methods

Visually impaired and blind individuals often encounter challenges in locating stores on the street, requiring guidance to ensure their daily activities are safe and effective. The objective of this study is to develop an assisted navigation system based on real-time semantic segmentation and Chinese text recognition. The proposed system aims to assist visually impaired individuals in obtaining information about the relative location of stores. It is highly user-friendly and can provide coordinate references in unfamiliar environments, aiding them in various activities like navigation and purchasing. In this work, we utilize the BiseNet v2 network, a real-time semantic segmentation model, to identify and detect shop sign objects. Additionally, we employ PP-OCR v3, an ultra-lightweight Optical Character Recognition (OCR) network, for Chinese text recognition. Furthermore, the system incorporates a speech broadcast function to convert visual information into auditory feedback. By providing real -time support for visually impaired individuals, this assistive navigation application demonstrates great potential in promoting public welfare.