Scene Text Detection And Recognition System For Visually Impaired People

Visually impaired(Ⅵ)people around the world have difficulties in socializing due to the limitation of traditional auxiliary tools.Scene text detection and recognition allow visually impaired people to obtain text information from outside which could profoundly change their life.However,text information in real life features complex background,low resolution,variable fonts as well as irregular arrangement,causing scene text detection and recognition technology quite difficult to realize.Based on Deep Neural Network,scene text detection and recognition system has proposed to help Ⅵ people.Firstly,with the help of Default Box and multi-scale prediction,regions of interest(such as bus,book,horizontal text)recognition network which could be trained end-to-end is designed to predict category and bounding box of the objects.Secondly,aiming at the difficulty of slanted text boxes,text detection network designs oriented rectangles regression scheme to improve the accuracy of special text box detection.Finally,text recognition dealing with Chinese and English hybrid characters deeply learned the texture features.It uses context feature to recognize text with variable lengths.Experiments on resource and precision tradeoff show that algorithms in this paper achieve competitive test results on standard datasets.To solve the problems existing in real scenes,multi-frame fusion technology augments features before text recognition,and lightweight neural network fights limitation of computing resources and real-time requirement.In general,algorithms designed in this paper suppress the influence of spatial transformation(perspective,scaling,rotation,etc.)and provide good accuracy and robustness for scene text description.The intelligent wearable device embedded above-mentioned algorithms and broadcast results through voice to pursue abundant scene description and excellent adaptability of different scenes.It could detect and describe route number of a coming bus,name of a book,brand name of a shop and other scene text content and background of it.This text detection and recognition system could combine other visual assistant algorithms such as navigation and obstacle avoidance to provide more accurate and efficient indoor and outdoor travel assistance for VI people.