Design And Implementation Of Embedded Wearable Reading-assistant System

Character is one of the main information carriers in the world.It carries the human civilization and spreads knowledge to people.Also,it is a culture promoter and history recorder.With the development of electronic information technology,one-dimensional code,two-dimensional code and color bar-code which can carry more information come out as the new information carriers.Among them,two-dimensional code is mostly used in daily life.However,it is still inconvenient for the blind and weak-sighted people to get information from texture or bar-codes.That is because the majority of current character recognition and bar-code recognition tools are scanning APPs on smart phones.The scanning will only be completed after launching the scanning APP and aiming the camera at the object.These operations are very difficult for blind and weak-sighted people.And the recognition feedback are always in text format presenting to users.So the reading problems of blind and weak-sighted people is still to-be solved.Therefore,it is necessary to make a reading-assistant device for the blind and weak-sighted people.This paper focuses on designing a reading-assistant device specifically for blind and weak-sighted people,intergrated with the popular wearable ideas.This paper makes several modifications and optimizations in hardware and device functions comparing to existing character and bar-code recognition products.And a wearable reading-assistant device specifically for blind and weak-sighted people is integrated with both character recognition and two-dimensional bar-code recognition functions.The main contents and innovations of this paper are as follows:1.This paper proposes a method to realize two-dimensional code decoding on an embedded platform with relatively few resources.Complete image capturing and storing with special Bit-Band region and direct memory access double buffer mechanism.Put forward a connected-region tracking method to locate the position pattern in QR code,thus the floating point operations will be saved.Get sampling image on the current rotation angle without rotating the original image.After that,two-dimensional bit stream can be decoded on an embedded platform with relatively few resources.2.Use Stroke Width Transform algorithm to detect the text area in an image.Preserve original text information as much as possible.3.Distinguish multi-lingual characters with their own features.Such as minimum circumscribed rectangle height-width ratio,stroke width ratio or characters’ gap.4.Adapt semi-supervised learning in training alphabet and Arabic numbers.Use Google open-source recognition engine Tesseract-OCR to recognize Simipified Chinese characters.Experiments show that the training of ANN takes much time but it takes very little time in predicting.Tesseract-OCR also has a very good Simplified Chinese characters recognition rate.5.Design a small and low-cost wearable reading-assistant device with image recognition function on an embedded platform with relatively few resources.The main functions of the newly designed device include:Two-dimensional bar-code offline decoding on an embedded platform;Characters online recognition on a server through WiFi.The device will send the two-dimensional bar-code decoding result to TTS(Text to Speech)module or send the character recognition result in GB2312 format through WiFi to TTS module.So that the blind and weak-sighted people can hear the original information in text or bar-code.The whole design is a wrist watch-sized wearable device with low power comsumption.The reading-assistant device designed in this paper has low power consumption and good recognition rate,also,it is convenient to use by blind and weak-sighted people.The device can not only serve the blind and weak-sighted people,but also help children and language learners to learn a foreign language.Part of the hardware design and code is pushed to GitHub for open source learning.