Research on Vectorization Method of Architectural Floor Plan Images

Rong Li1, Kun Zou1, 2, *,Wensheng Li1, 2, Liang Yang1 1University of Electronic Science and Technology of China, Zhongshan Institute, Zhongshan, 528402, China

2School of Computer Science andEngineering, University of Electronic Science and Technology of China, Chengdu, 611731, China

* Corresponding Author: Kun Zou

Keywords: Morphological, Threshold segmentation, Contour detection, Vector, Hough transform

Abstract. A method of automatic vectorization is proposed for nonstandard architectural floor plan images. It contains four steps: image preprocessing, segmentation, vector, classification. Use different algorithms in different regions. Classify the lines after vectorization according to geometric characteristics and prior knowledge. The result of multiple experiments has testified the feasibility and efficiency of the algorism. The method can be used in furniture customization, home decoration display, virtual reality and other application fields.

Summary

With the development of computer science, graph recognition technology has been widely used in many types of technical documents and drawings analysis. But the research on automatic analysis and processing of construction drawing is still relatively few.

Although the scientists at home and abroad continue to improve the vectorization of raster images and vector method, the existing vectorization software level is not very high, mainly reflected in the recognition processing speed, anti-noise, various characters and unsatisfying curve. For architectural drawings, most of the objects in the current system are the standard architectural design, not the floor plans after prettification in the sale of the housings.

Aiming at the demands of furniture customization, home display and other application areas, this paper achieve rapid automation, vectorization and classification based on the common housing units map. The image requirements are relatively low and the architectural design is unprofessional. The input can be the floor plan images in the publicity materials when people purchase the housing.

Overall Scheme

The early architectural floor images are generally drawn on paper, which can be transformed into images by digital sweeping instruments. At present, floor images can be easily obtained from related channels. Images are the most available and accessible to ordinary people. According to the vectorization of these images, the results can be imported into the 3D modeling software to generate three-dimensional apartment drawings, which can be convenient for home decoration design or furniture customization.

Image Pretreatment

The following tasks are required to fulfill in the image preprocessing stage, such as size constraint, image enhancement, noise removal and so on.

First, the image can be partly cutting. This process requires manual intervention. But, due to the simple operation and quick speed of the cutting, it does not increase the difficulty of operation. We can quickly remove a lot of non-building component content, such as real name, advertising words and so on.

2017 5th International Civil Engineering, Architecture and Machinery Conference(ICEAMC 2017)

Copyright © (2017) Francis Academic Press , UK 31

Then, thsize. To de

Finally,After do

operations

Image Seg

Obtainmapartment access the stage. Theroverall out

In termpretreatmegenerally c

In termsTo obtain apartment,satisfy the

src(seedsrc(seedsrc(seedWall In

categories:circular arcsolid and sbelongs to the apartmreconstructarea. SomeDifferent w

Figure 1

he size conseal with the image syntoing the abto remove

gmentation

ment of Exlayout map apartment lrefore, we wtline in the f

ms of the grent. The excontained ins of the colothe overall and then ufollowing c

d.x,seed.y)r-d.x,seed.y)gd.x,seed.y)bnspection. A: fine structuc, and text oolid circle. the area, otent layout, ttion effect ie of the imwall forms a

1. Threshold

straints shoulater segmethesis is donove treatmesome noise

n and Vecto

xternal Ovcontains so

layout diagrwill focus thfirst step. ray image,

xternal contn this profileor images, thcontour of

use the regiconditions: -lo_diffr<=

g-lo_diffg<=b-lo_diffb<=According ture and bloobject; the laIf the width

therwise it ithe segmentis correct. Tages are roare suitable

d segmentat

uld be takenentation andne to achievent, we use .

rization

erall Outliome other corams proceshe apartmen

we shouldtour integrae. here is usuathe apartmion growing

src(x,y)r<==src(x,y)g<==src(x,y)b<=to the shapeck structureatter categoh of the objes a thin wirtation and d

The wall shaugh, some to different

tion Figure

n. Differentd recognitiove the goal oe the approp

ine. In addiontent. Espessed and beant layout ar

d conduct tal, the cont

ally a color ment, the firs

g algorithm

=src(seed.x,s=src(seed.x=src(seed.x

e, the apartme. The formory includesect in the imre structure. detection resape is thick images of

t detection m

2. Outline d

t image souon, the sizes of enhancempriate size o

tion to the ecially, the oatified by threa and exte

the binarizatour inform

or texture inst step is to

m. The colo

seed.y)r+upx,seed.y)g+ux,seed.y)b+ument layout er category a variety of

mage is greaAs one of t

sult of the wline entitiethe wall aremethods.

detection Fi

urces often lshould be u

ment. of the kerne

relevant eleordinary homhe advertisinernal contou

ation and cmation wind

n addition topartition th

or values of

p_diffr and up_diffg andup_diffb. drawings ccontains a

f specific syater than a cethe most imwall directlys in some ime drawn co

igure 3. Out

lead to diffeunified first

el to do mo

ement inforome buyers ang companyur to acquir

contour detedow and th

o the type inhe backgrouf pixels in f

d

can be dividvariety of t

ymbols suchertain value

mportant comy related to tmage, belon

ompletely in

tline filterin

erent imaget.

rphological

rmation, theare prone toy in the latere the entire

ection afterhe door are

nformation.und and thefixed range

ded into twothin, hollowh as arrows,e, the objectmponents ofthe final 3Dnging to then thin lines.

ng

e

l

e o e e

r e

. e e

o w , t f

D e .

32

The vecFigure 5 shshows the t

The vec

Inspectand the outused to finthe inside aThe outlineof the walldetection oFigure 10. obtain the longest percontour. Tindicates thyellow line

01

01

01

01

1

ctorization phows the orthin lines pa

Figure 5. octorization o

tions of theter contour

nd out the loand outsidee of the wall is enlargeof the outer We examicategories

rimeter. Thehe rest are he wall; thee indicates t

0000

0000

0000

0000

1111

process of iginal floor art of the flo

riginal flooof architectu

e Door, the line, the fea

ocation of the contours. Tll and externed to avoid

contour in ine the contof doors, b

e basis of ththe outline green line rthe entrance

0

0

0

0

1

00

00

00

00

11

Figure 4.the composplan; Figur

oor plan.

r plan Figurural floor pl

Figure 8. LWindow a

atures of thehe balcony, The specificnal contoursthe possibilFigure 9. Wtours of Figbalconies a

he discriminof the windrepresents te door.

1000

1000

1000

1000

1111

Convolutiosite wall isre 6 shows t

re 6. Thick lan drawn b

Line segmenand the Bale exterior cwindow an

c process is s are drawnlity of a litt

We subtract gure 10, an

and windownation gate isdow, and ththe window

11

01

01

01

01

on kernels shown in

the thick lin

lines part Fby the thin li

nt detectionlcony.Basedontours of td entrance dshown in F

n into the samtle deviatiothe wall bo

nd then clasws. The bass the numbehe result is ; the blue li

111

000

000

000

000

Figure 5, Fnes part of th

igure 7. Thiines is show

n d on the detthe windowdoor with th

Figure 9, Figme image inn in the ed

old drawingssify the resis for judgier of the conshown in Fne represen

11

00

00

00

00

Figure 6 anthe floor pla

in lines partwn inf Figur

tection of thw, balcony ahe differencgure 10 andn Figure 9.

dge positiong from Figursulting conting the balncentrated pFigure 11. Tnts the balco

111

100

100

100

100

d Figure 7.an; Figure 7

t re 8.

he wall linend door are

ces betweend Figure 11.The outline

n during there 10 to gettour sets tocony is the

points of theThe red lineony, and the

. 7

e e n . e e t o e e e e

33

Conclusio

We adoalgorithm athe paper. vectorized Figure 13.double-lineXML file,

Fig

ns

opted the Vand teste thIn terms oand the wa We can ae without fuand achieve

gure 9. Com

10. Differ

Visual Studhe algorithmof the solidalls, windowlso correctl

furniture. We the correc

mbination of

rence betwe

Figure

dio 2013 anm through md line wallsws, balconiely identify

We save the ct result in th

f internal an

een internal

11. Detectio

nd opencv3many images, most of ts and internand vectoricoordinate

he later 3D

nd external c

and externa

on result

3.0 in the ps of differenthe images nal doors caized the floinformatioreconstruct

contours Fig

al contours

personal cont shapes ancan be corn be identif

oor plan dran after vect

tion.

gure

omputer to nd differentrrectly segmfied, which awn by sol

ctorization,

realize thet sources inmented andis shown inlid-line andsuch as the

e n d n d e

34

The alg

improvemeto improvedistortion a

Throughvectorizatiselect the twalls, doodifferent st

Acknowle

This resalso the reGugangdonZhongshan

Reference

[1] Zhou JVectorizatiEdition), 1

orithm can ent of the ale the robustnand enhanceh the autheon for aparttype of imagors and wintyles quickl

dgement

search was esult of Culng Provincn City (Gran

es

Jiliu, Guo ion System 998, 35(1):

realize the lgorithm to iness of the ae the efficie

entication oftment layouge recognitindows. We y and accur

funded by Nltivation Ple (Grant Nnt No. 2017

Jing, Zhanof Enginee

65-70.

Figure 12. S

identificatioidentify the algorithm toency is the df many imaut drawing. ion. We cancan achiev

rately to lay

National Nlan of Exce

No. YQ2015B1115).

ng Ye, Zhuering Drawi

Some exper

on of housee internal doo adapt to thdirection of ages, we coWe can co

n also providve the diffey foundation

Natural Scienellent Youn5241) and P

u Hui, Longings [J].Jou

rimental res

ehold doors or is the nex

he apartmenf our efforts.onclude thatnsider all pde the choicerent methons for the 3D

nce Foundang TeachersPlan Projec

g Jianzhongurnal of Sich

sults

and windowxt step of th

nt layout ma. t there is no

possibilities ce of the styods of vectoD reconstruc

ation (Grants of Univerct of Scienc

g, Chen Zhhuan Unive

ws in the bahe research. ap with pers

no fixed meand allow t

yle, such as orization acction of the

t No. 61502rsities and Cce and Tec

hongling. Aersity (Natu

alcony. TheIn addition,pective and

thod of thethe users tothe style of

ccording toe next step.

288). It wasColleges in

chnology of

A Study onural Science

e ,

d

e o f o

s n f

n e

35

[2] Ding Weidong, Yuan Jingqi. Study on Vectorization Algorithm of Engineering Drawing [J]. Computer Engineering, 2006, 32(24): 157-158+161.

[3] Lu Zongqi, Zhang Qiuping. Line Contour Tracking in EngineeringDrawing Vectorization [J]. Journal of Image a nd Graphics, 1997, 2(12): 878-882.

[4] Zhang Qing, Guo Wei. Depth images-based approach for 3D reconstruction [J].Journal of Huazhong University of Science and Technology (Natural Science Edition), 2008, 36(2): 88-91.

[5] Lu Guodong, Ruan Jianzhong, Jiang Tao, Peng Qunsheng. Research on 3D Reconstruction of Objects from Engineering Drawing [J].Journal of Computer Aided Design andComputer Graphics, 1999, 11(6): 516-520.

[6] Yin X, Wonka P, Razdan A. Generating 3D building models from architectural drawings: a survey [J]. IEEE Computer Graphics & Applications, 2009, 29(1) :20.

[7] Dori D, Liu W. Sparse Pixel Vectorization: An Algorithm and Its Performance Evaluation [J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 1999, 21(3): 202-215.

[8] Lu T, Tai C L, Yang H, et al. A Novel Knowledge-Based System for Interpreting Complex Engineering Drawings: Theory, Representation, and Implementation [J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2009, 31(8):1444-1457.

36