Research On The Recognition And Vectorization Algorithm Of Grating Floor Plan Image

Raster floor plan refers to the raster image used to describe the layout of the floor plan.In the virtual decoration of the house,users participate in th e layout and design of the room,not only can they apply their decoration ideas to their own home decoratio n,but also allow users to obtain the most suitable apartment configuration.Traditional house model generation mainly according to the designer manually adding the house structure according to the two-dimensional map of the house to obtain the vectorized data of the house map,and then performing three-dimensional reconstruction based on the prior knowledge in the height direction to obtain the three-dimensional house model.Because of internal structure is complex and contains many component elements,manual extraction of vectorized data is not only time-consuming and labor-intensive,but also inefficient.In order to solve the problems of long modeling period and many manual interactions in the virtual decoration display of the current apartment type,thi s paper has designed a set of technical routes for the recognition of the floor plan,which realizes the segmentation and recognition of the main structural elements in the floor plan and converts it into vectorized data required for 3D modeling.The main work of this article is as follows:(1)A recognition algorithm based on multi-attribute analysis is proposed for the structural elements that are common in the floor plan,which realizes the recognition of scales,walls,bay windows and columns in the floor plan.In this paper,the template matching method is first used to retrieve the size identification area from the figure,and the scale is obtained by identifying the number of the dimension label and calculating the distance between the end points of the scale.At the time,the wall area is extracted according to the feature with the highest wall color proportion in the outline of the floor plan,which solves the problem of difficulty in detection due to different wall colors in the floor plan recognition,and at the same time,the thinning line algorithm is optimized so that the wall can retain the angular relationship between the walls when extracting the centerline;finally,according to the morphological characteristics of the bay window and the pillar,the bay window recognition algorithm based on the bump feature and the pillar recognition algorithm based on morphological features.(2)A door and window recognition algorithm based on deep learning is proposed to realize the classification of doors and windows.Because the doors and windows are embedded in the wall,they are all composed of a rectangular frame and multiple short lines in shape.The difference is that the door has more arcs and vertical lines than the window.The accuracy rate is low.After extracting the sub-pictures containing doors and windows,this paper uses the deep classification ability of deep learning to identify them,which improves the accuracy of door and window recognition.These include the design and use of annotation tools,the construction of door and window data sets,the determination of door and window candidate regions,and door and window classification based on deep convolutional neural networks.(3)Build a floor plan recognition system.It mainly includes data inp ut module,algorithm processing module,human-computer interaction module,and data output module;the system after adding manual interaction module can allow users to supervise the identification of the system,and recognition errors can be seen in the v isual recognition result chart to allow designers to make corrections quickly.