Research On Floorplan Recognition For Automatic Decoration Of Building Interior

With the continuous improvement of living standards,people's demands for the home environment have also been continuously improved,which has led to a huge home improvement market,and many home improvement companies have emerged.Renovation program design ability is the core competitiveness of home improvement companies,but the manual design methods commonly used today have problems with long design cycles and high costs.Therefore,the market urgently needs an automatic/semi-automatic home improvement design system.Automatically extracting the range structure and size information from the floor plan is an important step to achieve automatic/semi-automatic home improvement design.This is exactly the problem that this article studies.The main work of this article is as follows:(1)House plan scale recognition:scale is an important basis for restoration of house size.This dissertation calculates the scale by means of scale point positioning,size standard text positioning,and dimension text recognition.For the problems that the scale positioning may be inaccurate and the character recognition may be wrong,this dissertation makes use of the redundancy of dimensional standard information,proposes to use the voting principle to correct the scale recognition results,and further improve the accuracy of scale recognition.(2)Based on the prior knowledge of the wall detection:Wall is the basic element of the floor plan.This dissertation first uses the characteristics of the wall,such as the presence of edges on both sides and the uniform internal color distribution,and the self-similar characteristics of the wall to detect the preliminary candidate wall.Then,the wall structure in the user's profile is then used to further optimize the preliminary recognition results,obtain accurate wall alignment results,and accurately connect the wall recognition results to lay a good foundation for subsequent processing.(3)Door and window recognition based on deep learning:In the floor plan,doors and windows are represented by related symbols.Because of the positional relationship between the door,window and wall,this dissertation completes the identification of door and window elements according to the steps of"determine the position of doors and windows,then classify doors and windows".That is,sub-graphs containing doors and windows are extracted first,and then image classification is performed.Due to the variety of floor plan styles and the different windows and doors,this dissertation use the powerful classification ability of deep learning to realize door and window recognition.