Architecture Atlas Retrieval System Based On OCR Recognition Technology

Building atlas is a common practice designed,tested or operated by technicians to avoid repetitive labor,and it is an important part of engineering construction standardization.When designers design engineering drawings,if they need to use the contents of the building atlas,they only need to mark the atlas number on the drawings to simplify the drawing representation.When the construction workers need to refer to the building atlas on the construction site,or look up the paper version or search for the electronic version,it is neither convenient nor efficient.With the development of mobile Internet technology,the use of intelligent methods makes it convenient and efficient for building engineers to query the atlas,which has been widely concerned.Intelligent identification and Atlas retrieval of building atlas number on the mobile end has a high research value and application prospect.The system studied in this dissertation is a development project which applies OCR recognition technology to the field of architecture.The goal of recognition is the number of atlas on architectural drawings.Because the architectural drawings are mostly blue prints and the font is also blue,the foreground information and background information are not easy to distinguish,and the drawings will be damaged and faded after long-term use by the construction workers.In addition to the special coding rules of atlas number,there are problems such as circle,split line,upper and lower digital coding,and the atlas number is interfered by other graphics and numbers around the construction drawings,The OCR technology can not be used directly,which also brings great challenges to the location and recognition of atlas number.Aiming at the above problems and combining with the practical work needs of engineering professionals,this paper proposes an OCR recognition technology based on deep learning and image preprocessing,which extends the application of OCR technology to a new field.The main work of this dissertation includes the following aspects:(1)System architecture design.Through in-depth research and analysis of user requirements,a large number of random characters around the atlas number in the drawing and traditional positioning of structure will generate a large number of redundant information,and the circular features in the atlas number will cause inaccurate recognition,the system is designed.The functions of the system are mainly divided into six modules,including user login and registration,image acquisition,Atlas number detection,Atlas number segmentation,Atlas number recognition,building atlas retrieval.(2)Algorithm selection and technology adoption of system implementation.The popular algorithms in key modules such as Atlas number detection,recognition and Atlas retrieval are fully compared by experiments.Finally,the end-to-end object detection algorithm yolov3 is selected,and a circle positioning and cutting algorithm based on the maximum contour is proposed.Baidu OCR character recognition technology is adopted and Alibaba cloud OSS is used as the storage service.(3)System function realization.The system uses the client / server structure to distribute the tasks to the client and the server reasonably.The client uses the Android front-end development technology,and the server uses the flask micro framework written by Python to develop.All the module functions of the system are realized.The experimental test shows that the whole system has achieved the expected effect of engineering application.The application of the system provides convenience for engineers to query atlas and greatly improves work efficiency.We can also apply the content of this paper to the AR glasses and other intelligent devices,so as to really liberate our hands.Therefore,the system has a good application value and development prospects in the field of architecture.