Extracting Method Of Structure From Floorplan Image Based On Deep Learning

In recent years,the continuous development of virtual reality(virtual reality,VR)has injected new vitality into the home decoration and real estate industries.A number of online home decoration design companies have emerged,and VR online house viewing has also been favored by real estate agents.The traditional method to obtain the three-dimensional house model is generally designed by software such as 3d Max,or by manually drawing the vector house structure from the two-dimensional plan of the floorplan images,and then generating the model.However,these methods are timeconsuming and have low efficiency,which is not conducive to rapid model generation.Therefore,it is particularly important to extract the house structure from the floorplan images automatically and quickly.There are various styles of floorplan images on the market without unified standard.At the same time,the floorplan images also contain a lot of decorative furniture,which increases the difficulty of extraction.This paper designs a general method for the difficulty of extracting the structure of twodimensional floorplan images.The main research contents are as follows:(1)Wall edge extraction based on wireframe.Wireframe extraction is to obtain the prominent straight lines in the image and the intersection points between the straight lines.There is a non-single color in the wall part of the house in the floor plan,but the edge information is a universal feature.This article proposes to add constraints to the network according to the characteristics of the wall edge line which is generally horizontal or vertical,and find the line segments at the edge of the wall,then find the parallel line group and extract the center line as the vectorization result of the wall.(2)Extraction of doors,windows and scales based on target detection.Target detection is a type of research task in deep learning.Through the learning of labeled datasets,similar targets in the pictures to be detected can be identified.In this article,target detection is used to obtain the positions of doors,windows and scales.According to the center line of the wall,we optimize the position of the doors and windows,and optimize their length according to the canny edge map.(3)Scale identification.Scale recognition is identifying the scale lines and their corresponding labeled numbers.We use template matching to identify the numbers because their shapes are similar.Finally,the interference results are eliminated by clustering to improve the recognition accuracy of the scale.(4)This article evaluates the method in two perspectives: quantitative and qualitative.We use line segment matching to determine the correctness of the detected wall,door,and window objects,and compare with my previous research methods.For qualitative assessment,we compare our result with two existing commercial softwares,and analyzes their advantages and disadvantages respectively.The quantitative and qualitative experimental results show that the method is effective in extracting structures from different types of floorplan images.