Research On Intelligent Recognition Technology Of Building Floor Plan Based On Deep Learning

With the improvement of living standards,people’s requirements for the quality and efficiency of interior design are increasing,which promotes the interior design industry to step forward to a higher level of intelligence.Identifying the geometric information of building components and furniture positions in building floor plan pictures and creating virtual scenes according to the recognition results are the key to the intelligence of interior design.The existing automatic recognition methods is not accurate so it have to rely on manual recognition,which makes the generation efficiency of three-dimensional scene is difficult to satisfy the users.In this paper,a method to identify the building floor plan automatically based on deep learning target detection is studied,and complete intelligent recognition algorithm method is realized based on the professional interior design platform.The main contents of this paper are as follows:(1)In this paper,recognition method for building floor plan based on deep learning target detection technology is proposed.By comparing the performance of different target detection algorithms in recognizing building floor plan,Center Net algorithm is selected as the basic algorithm.Feature fusion technology and attention mechanism is applied to the feature extraction network to enhance the effective information content in the detection feature map;distance Io U loss(DIo U)is introduced to improve the loss function and further improve the positioning accuracy.The improved algorithm reaches 91% of mean average precision for the detection of building components,and the experimental results verify the feasibility of target detection technology in this field.(2)In this paper,a method to quickly extract the vectorized geometric data of building components is studied.First,the detection model is used to recognize the picture,then according to the geometric position relationship between the bounding boxes of the building components,the vector lines are quickly extracted and the wall connection points are fused to obtain the vector data which can be used to accurately reconstruct the vectorized floor plan structure.(3)In this paper,an recognition method for layout feather based on deep learning is studied.The target detection model is used to identify the furniture in the building floor plan and the rooms in the corresponding vectorized floor plan is laid out automatically according to the detection results;if there is no furniture element in the picture,The long short term memory(LSTM)network is used to identify the characteristics of floor plan section and predict the layout results.Classifier based on deep learning is trained to recognize the decoration style of a selected indoor effect drawing,and automatically select the same kind of furniture model for room layout.(4)The intelligent floor plan recognition system is realized on the upper layer of V-life interior design software platform.The system is divided into structure identification and reconstruction module,furniture selection module and layout module.Different modules provide independent interfaces to realize the corresponding functions.Tested on the platform of V-life software,the results show that methods realized in this paper can recognize the building floor plan quickly and accurately,which accelerates the intelligent generation of three-dimensional floor plan scenes in interior design software.