導盲系統的架構，經常搭配視覺偵測與前導機器人應用，來作為障礙物的偵測與行徑路線規劃，本研究提出一個可以增加視覺資料應用性的招牌文字辨識系統，使收集來的影像資料，可以達到更好的應用效果，讓影像不只可以偵測障礙物，同時還可藉由影像上的文字訊息，得知周遭的環境狀態。首先會將街景影像中的招牌圖片切割出來，使文字偵測的區塊可以縮小，並針對招牌文字使用CRAFT 文字偵測模型，找出圖片中擁有文字像素的範圍，有了文字區塊圖片，就可以運用Tesseract 文字辨識模型讀出文字的內容，最後得出的文字資料，可交給電腦或微控制器使用，並結合導盲系統的應用，使盲人朋友們可以清楚知道周圍有哪些商家。;The architecture of a guide system, often in combination with visual detection and leading robot applications, as a barrier detection and route planning. This study proposed a signboard text recognition system that would enhance the application of visual data. The system would enable the collection of video data to achieve better application results so that the image could not only detect obstacles but also use text messages on the images to identify the surrounding environment. First, you can cut out the picture of a street scene so that the block of text detection can be reduced. Then, the CRAFT text detection model is used to find out the range of text pixels in the picture. With the image, the Tesseract text recognition model is applied to read the text. The final textual information can be used by computers or microcontrollers, combined with the architecture of a guide system to improve the identification of the text messages, so those blind friends can clearly know which shops are around them.
