基于视觉的错序在架图书检测系统

A Vision-Based System to Detect Books with Incorrect Sequence on Shelf

在架图书整理是图书馆的一项常见工作,迫切需要提高自动化水平。本文以尽可能不增加额外投资和设施改造为前提,提出了依靠移动设备获取图像,通过视觉技术构建软件系统以实现对错序图书的自动识别的方法。该方法由移动设备采集在架图书图像,利用LSD算法检测书脊边缘并分割图书图像,进而得到索书号区域,并通过投影分离出索书号区域中的字符。采用前馈连续双卷积CNN网络结构模型并综合使用了Dropout、Re LU激活函数和Softmax回归等技术完成对字符的识别。最后,将组合后的索书号与数据库预存信息进行比较,得到图书的正确排列信息。对系统的实际测试和分析表明,本文的方法可以针对一般索书号在架图书进行较好的错序检测,能为图书的自动化整理提供一种可行的方案。

Sorting books is a regular task of libraries,and it＇s urgent to improve the automation level.Without any additional investment and infrastructure improvement,an approach to detect books with incorrect sequence on shelf by means of a software system based on vision inspection technique is proposed.Firstly,mobile devices are employed to capture images of books on shelf.Secondly,LSD algorithm is used to fnd all edge lines of book spines,and characters of call number are able to be gained by means of projection from ROI separated from a book image.Thirdly,a feed forward continuous double convolution network model is built to perform the task of character recognition,where other techniques such as Dropout,ReLU activation function and Softmax regression are also included.Finally,the reconstituted call numbers are compared with and the recorded information in a database to obtain correct positions.Experiments and analysis by running the system show that the approach is able to ensure good performance on fnding those books with incorrect sequence on the shelf.