Research On Floorplan Image Recognition Based On Shape And Edge Features

Floorplan design is an indispensable part of house construction,and floorplan image is the direct result of floorplan design,which is an important medium for purchaser and builder.At present,it's an inevitable trend for three-dimensional technology based on virtual reality,to replace the traditional two-dimensional floorplan image.Moreover,the technology has been successfully applied into market aboard,but just start internally.In addition to expensive equipment,the most barriers for the technology to be widely used,is that the 3D models are pre-constructed and the number is limited.In order to solve the problem that the current three-dimensional virtual display device can't automatically analyze the floorplan image,this paper designs a set of technical route about floorplan image recognition,and build up the software and hardware platform to carry out the algorithm research.According to the recognition target and mission requirements,this paper proposes a recognition scheme based on CMOS camera.Through the analysis of the image quality captured by Cannon camera,a mathematical model is created to analyze camera imaging and lens distortion.Combined with the recognition of dimension,we recover the real size of floorplan.After the observation and analysis of a large number of floorplan,we select the shape and edge feature as the feature of recognition.Based on the shape feature,we design segmentation and recognition algorithms,which include wall segmentation,correction,and topology analysis to complete the identification of the main elements in floorplan image.Finally,we create the corresponding 3D model by stretching and embedding.Based on edge feature,we design the pixel edge extraction,sub-pixel edge extraction,vectorization and wall identification algorithms to realize the automatic segmentation of the main elements in floorplan image,and highly precise transformation from the bitmap to vector.Through experiments on a large number of units,we prove that the recognition accuracy of the algorithm is high,and the robustness is good.