嵌入式可穿戴阅读辅助系统设计与实现

发布时间：2018-06-15 09:07

本文选题：嵌入式平台 + 阅读辅助　；参考：《华东师范大学》2017年硕士论文

【摘要】：在现实生活中,文字作为信息的主要载体形式之一,承载着人类文明,向人们传播知识,弘扬文化,记录历史。随着电子信息技术的发展,出现了一维码、二维码以及彩色条形码等能够包含较多信息的条码图像作为信息载体,其中二维条码是主要的信息载体。然而,对于盲人和弱视力者而言,在获取文本信息或条形码信息时,仍存在诸多不便。这是因为目前市面上大多数文字识别或者条形码识别的工具是智能手机端的APP,使用时需要打开APP,并将摄像头对准文本区域或者条码区域。对于盲人和弱视力者而言,使用难度较高,且多数智能手机APP的识别结果仍是以文本的形式呈献给用户,所以并没有从根本上解决盲人和弱视力群体的阅读问题。因此,研究一款专门为盲人和弱视力者进行阅读辅助的设备就显得非常必要。本文旨在融合现有的可穿戴设备理念,设计一款专门为方便盲人和弱视力者使用的可穿戴阅读辅助系统,在研究和比对现有的文字识别或条码识别产品的基础上,对其功能和实现平台进行改进,结合无线网络,设计一款能实现文字识别和二维码识别的可穿戴设备,为盲人和弱视力者进行阅读辅助。本文主要工作内容及创新点如下:1、提出一种在资源相对较少的嵌入式平台下实现二维码译码的方法。通过本文选择的芯片所特有的位带存储区以及DMA(直接内存访问)双缓冲机制,快速完成图像采集、二值化和压缩存储;提出使用连通边界跟踪算法寻找二维码定位符,减少大量浮点运算;不对图像进行旋转,直接在当前旋转角度上进行采样,进而获取二维码比特流信息,实现二维码在资源相对较少的嵌入式平台下的译码。2、通过提取图像中各像素笔画宽度特征检测图像中的文本区域,较大程度地保留文本区域原始信息,检测效果较好。3、根据不同语言的字符差异,分析并提出利用字符分割时得到的最小外接矩形高度与宽度的比值、字符笔画宽度和字符间距等特点区分不同语言字符。4、使用半监督学习方式训练神经网络识别英文和阿拉伯数字,用Tesseract-OCR开源引擎识别中文简体汉字。实验发现,半监督学习算法训练英文字母和阿拉伯数字所花的时间较长,但训练好后,识别效果较准,速度较快;Tesseract-OCR也具有较好的汉字识别效果。5、设计了一款小体积、低成本的嵌入式可穿戴阅读辅助设备。在资源相对较少的嵌入式平台上实现图像识别、网络通信、语音朗读等功能。具体实现的功能主要包括,在嵌入式平台上实现二维码离线译码;通过无线网络,在服务器端实现文字在线识别。将嵌入式平台下二维码译码结果语音朗读给盲人或者弱视力使用者,也可将服务器端文字识别结果的GB2312编码通过无线网络发送至嵌入式平台的语音模块朗读。硬件设计上充分考虑设备的可穿戴性、低功耗、散热性等问题,设计出外观与手表相近的可穿戴设备,盲人和弱视力使用者仅需要通过简单的屏幕触摸即可阅读文字或识别二维码。本文所设计的可穿戴式阅读辅助系统,达到了成本较低,识别率较好,使用方便的效果。既方便盲人和弱视力的人阅读文献和二维条码,也方便语言学习者,儿童等人群的使用。部分代码和硬件设计方案已发布和托管到GitHub上。
[Abstract]:In real life, text is one of the main forms of information, carrying human civilization, spreading knowledge, carrying forward culture and recording history. With the development of electronic information technology, a bar code image, such as one dimension code, two-dimensional code and color bar code, which can contain more information, is used as the information carrier, in which two dimensional bar code is used. However, for the blind and weak eyesight, there is still a lot of inconvenience in obtaining text information or bar code information. This is because most of the text recognition or bar code identification tools on the market are APP on the smartphone end, and when they are used, the APP should be opened and the camera is directed to the text area or the text area. Bar code area. For the blind and weak eyesight, it is difficult to use, and the recognition results of most smart phone APP are still presented to the user in the form of text, so there is no fundamental solution to the reading problems of the blind and weak eyesight groups. Therefore, the study of a device for reading auxiliary for the blind and weak eyesight people is to be studied. The purpose of this paper is to integrate the existing wearable device concept and design a wearable reading assistant system for the convenience of the blind and weak eyesight. Based on the research and comparison of the existing text recognition or barcode recognition products, the function and implementation platform are improved, and a wireless network is designed to design a model. The wearable device that can realize word recognition and two-dimensional code recognition can be used for reading for the blind and weak eyesight. The main contents and innovations of this paper are as follows: 1, a method for the realization of two-dimensional code decoding is proposed in the embedded platform with relatively few resources. In memory access) double buffering mechanism, the image acquisition, two value and compression storage are completed quickly, and a connected boundary tracking algorithm is used to find two-dimensional code locator and reduce a large number of floating-point operations. The decoding.2 under the relatively small embedded platform is used to detect the text area in the image by extracting the feature of each pixel's stroke width in the image. The original information of the text area is preserved to a great extent, and the detection effect is better.3. According to the character difference of different languages, the minimum outer rectangular height obtained by the character segmentation is analyzed and proposed. The ratio of width, character stroke width and character spacing distinguish different language characters.4, use semi supervised learning method to train neural network to recognize English and Arabia numbers, and use Tesseract-OCR open source engine to identify Chinese Simplified Chinese characters. Experiments show that the time spent by semi supervised learning algorithm training English letters and Arabia numbers Long, but after training, the recognition effect is more accurate and the speed is faster; Tesseract-OCR also has a good Chinese character recognition effect.5. A small, low cost embedded wearable reading assistant is designed. The functions of image recognition, network communication and voice reading are realized on the embedded platform with relatively few resources. It mainly includes the off-line decoding of two-dimensional code on the embedded platform; the text online recognition is realized on the server side through the wireless network. The two dimensional code decoding results under the embedded platform are read aloud to the blind or weak eyesight users, and the GB2312 encoding of the server end recognition result can be sent to the embedded system through the wireless network. The hardware design takes full account of the wearability, low power consumption and heat dissipation of the equipment, and designs a wearable device with similar appearance to the watch. The blind and weak eyesight users only need to read the text or identify the two-dimensional code through a simple screen touch. The wearable reading assistance designed in this article is designed. The system has achieved a lower cost, better recognition rate and convenient use. It is convenient for the blind and weak eyesight to read the literature and two-dimensional bar code, and also facilitates the use of language learners and children. Part of the code and hardware design has been published and hosted on GitHub.
【学位授予单位】：华东师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.4;TP368.33

【参考文献】