基于小波分析和概率Hough变换的书脊视觉识别

针对无人图书馆中机器人利用视觉进行书脊识别问题展开研究。根据书脊图像本身较复杂、单本书轮廓难以提取的特点，提出了运用小波分析做书脊轮廓增强的图像预处理方案；通过实验比较了运用Sobel算子和Canny算子做书脊轮廓检测的效果，用累计概率霍夫变换法做书脊线段查找，用VC++和OpenCV开发了应用程序；借鉴模糊计算思想，针对厚、较厚、中等厚度、较薄、薄、混合等几种类型的书脊进行了分类实验和研究，再经辅助优化处理，基本可有效识别出不同情况下每本书的书脊轮廓，检测出每本书的厚度，为机械手利用视觉进行图书取放操作奠定了基础。

Book-spine recognition by robot vision based on wavelet analysis and probability Hough transform

Abstract:The problem of the book spine recognition by library robot vision is studied. According to the characteristics that the book spine image is complex and the outline of every book is difficult to be extracted., the image preprocessing idea is proposed, which uses Wavelet analysis to heighten the outlines of the book spine. Through experiments, Sobel and Canny edge detection methods are compared. And the Canny edge detection method is finally used to detect the book spine outlines. Probability Hough transform algorithm is adopted to find the lines of the book spine image. VC++ and OpenCV are used to develop the application program. According to the thickness (Very Thick, A Bit Thick, Medium Thick, A bit Thin, Thin, Hybrid) of a single book, several types of book spine recognition tests are performed. The outlines of every book spine are effectively recognized and the thickness of every book is obtained by programming optimization process. This study lays a foundation for the books grasp operating by manipulator vision.