Study on the locaton of buldng evacuaton ndcators based on eye trackng Yue L Tsnghua Unversty yue-l5@malstsnghuaeducn Png hang Tsnghua Unversty zhangp@malstsnghuaeducn Hu hang Tsnghua Unversty, zhhu@tsnghuaeducn ABSTRACT Evacuaton ndcator s an mtant factor affectng evacuaton effcency durng emergency evacuaton In order to study the relatonshp between the dstrbuton of human s ponts of regard and evacuaton ndcators durng buldng evacuaton, ths paper carred out two groups of buldng emergency evacuaton experments, to get the data of human s ponts of regard, one of the expermenters was selected to put on eye tracker durng the experment Ths paper presents a method to preprocess the data of eye tracker and further analyzes the pretreatment results The results show that evacuaton ndcators n dfferent areas are sutable for dfferent placement, whch based on the poston dstrbuton of human s ponts of regard For example, evacuaton ndcators n the corrdor are sutable for hgher placement, and evacuaton ndcators n the starcase are sutable for lower placement CCS CONCEPTS Desgn and analyss of algorthms Mathematcal optmzaton Dscrete optmzaton Network optmzaton KEYWORDS buldng evacuaton experment; evacuaton ndcator; ponts of regard; eye trackng INTRODUCTION Wth the development of socety and the growth of urban populaton, to meet the ncreasng demands of people s lfe, many large composte buldngs have sprung up, and become common carrers of human actvtes The ncrease of sze and space, structure and functon complexty of buldng, and the number of Permsson to make dgtal or hard copes of all or part of ths work for personal or classroom use s granted wthout fee provded that copes are not made or dstrbuted for proft or commercal advantage and that copes bear ths notce and the full ctaton on the frst page Copyrghts for components of ths work owned by others than ACM must be honored Abstractng wth credt s permtted To copy otherwse, or republsh, to post on servers or to redstrbute to lsts, requres pror specfc permsson and/or a fee Request permssons from Permssons@acmorg 07 Assocaton for Computng Machnery ACM ISBN 978--503-593-6/7/$500 https://doorg/05/35653567 people nsde the buldng,all of them wll exacerbate the hazard of sudden events such as fres and terrorst attack, and wll make t more dffcult to evacuate safely Evacuaton ndcator plays an mtant and rreplaceable role n rapdly evacuaton, however wth the ncrease of the complexty of the buldng, the evacuaton ndcators gradually expose a lot of problems The eye tracker s an eye-trackng devce whch was frst used n psychologcal experments Wth the mprovement of lne-of-sght trackng technology, eye tracker s frequently utlzed n dverse feld hang sen et al[] extended eye tracker s applcaton nto sts psychology, studed the psychologcal cogntve characterstcs of athletes' decson-makng; L haqong et al[] and hang je et al[3], utlzed eye tracker to study the drver's vewpont dstrbuton to mprove drvng safety and traffc effcency hou xny[] combned landscape of cty parks and eye tracker to study the mprovement of landscape desgn At present, there are not many research on the combnaton of eye movement and evacuaton Wezhen[5] used eye tracker to study the mode of emergency evacuaton of large stadum audence; hou Xueyan[6] utlzed eye tracker to assst n mprovng the dentfcaton gudance of subway staton, to enhance the moble effcency of crowd n the subway; L lhua et al[7] poneered to use eye tracker n buldng evacuaton experments to study the ndvdual and small group behavor n emergency To study ratonalty and mprovement of the locaton of buldng evacuaton ndcator n dfferent evacuaton scenaros, ths paper carred out the evacuaton experments n day-scenaro and nghtscenaro, utlzed eye tracker to collect data Compared wth tradtonal smulaton, evacuaton experment can obtan human's relatvely true psychology and behavor data n emergency; eye tracker s a common tool n psychologcal experments whch usually used to analyze the dstrbuton of human s ponts of regard, the data obtaned s qute accurate EXPERIMENT DESIGN The experment was conducted on Aprl 6, 07 at Luqng Buldng of Tsnghua unversty Luqng Buldng has floors, each of them s exactly the same structure, ncludng elevators and starcases whch are respectvely located on eastern sde and western sde of the buldng, and the elevator s adjacent to the starwell The starcase totally has of 0 steps, each step s 0cm long, 5cm hgh, 30cm wde and 50cm long at the corner of the starcase The frst and second floor are 38 m hgh, and the
thrd floor to 0th floor are 5 m hgh Evacuaton nstructors are nstalled n the corrdor and starcase of each floor The structure dagram of Luqng Buldng s shown n Fg Fgure : Structure Dagram of Luqng Buldng of Tsnghua Unversty Ths experment smulated the real evacuaton scene, and the evacuaton ndcators are placed n dfferent areas of each floor of Luqng Buldng Evacuaton ndcator s placed on one sde of the wall n corrdor and the heght s close to the ground Two knds of sgns are placed n the starcase, ncludng floor number sgn at each layer, evacuaton ndcator located at the corner of each floor and s hghly close to the ground Intally the subjects all gathered n a room on the 0th floor, when the alarm went off, the subjects were quckly evacuated from the room to the starcase, and then evacuated from the 0th floor to the frst floor One of the subjects was selected to wear eye tracker to record the eye movement All of the subjects are freshmen from Socal Scences of Tsnghua Unversty They are all the frst tme to come to Luqng Buldng, and not famlar wth ts structure and evacuaton route In order to ensure the authentcty of the evacuaton experment, ths paper only conducted an experment respectvely n the day and evenng, so as to prevent the subjects from beng too famlar wth the constructon structure The start and end of the two experments s the same, and so as the expermental route In addton, the experments adopt reward and punshment measures, the frst person who arrves to the frst floor wll receve a monetary reward, and the person who arrves latest wll deduct the money The am s to motvate the subjects to evacuate quckly, arouse the subjects' enthusasm, and make the experment closer to the real evacuaton The experment scene s shown as Fg Fgure : Emergency Evacuaton Experment 3 RELATED WORK 3 Human s Eye Movement Informaton processng of human s eyes heavly rely on vson, about 80% to 90% of the external nformaton s obtaned through eyes Eye tracker s an eye-trackng devce based on the dgtal vdeo, t s basc prncple s to capture the eye mage n real tme by nfrared camera of hgh frequency samplng, and then obtan eye movement data through mage processng [5] The concept and parameter ndex of human eye movement sn t consstent defned n many documentatons, the followng s the defnton of hgh degree of recognton There are three man forms of eye movement: fxatons, saccades and smooth pursut [6-5] The fxaton behaves as a stay on the observed target, the stay generally lasts at least 00ms Human s eye s not absolutely statonary when fxng, the eyeball s constantly shakng slghtly to see the object, ts magntude s generally less than degree Majorty of the nformaton s acqured and processed only when t s beng watched The pont whch gaze produces s called fxaton pont or pont of regard The saccade s a rapd movement of the fxaton pont, and both eyes move smultaneously, the perspectve s to 0 degrees, the duraton s 30ms to 0ms, the maxmum speed s 00 degrees/s to 600 degrees/s Durng the perod of the saccades, the eyeball receves lttle nformaton, because the mage s movng too fast n the retna and the vsual threshold s elevated as the saccade happens Ths knd of pont s called saccade pont Smooth pursut s a slow, bnocular-movng eye movement form, the movement must have a slow movng target to be mplemented Ths knd of pont s called unconscous pont In addton, eye tracker also produces some nose ponts durng the operaton Therefore, the eye movement data collected by the eye tracker contan ths four knds of pont: the fxaton pont, the saccade pont, the unconscous pont, and the nose pont Only the fxaton ponts are vald data The other three knds don t contan vald nformaton, they are collectvely referred as the error ponts, and wll be deleted by data preprocessng 3 Eye Tracker s Orgnal Data The eye tracker brand used n ths paper s Yarbus, ts orgnal data fle ncludes the vdeo fle n mp format, and the detaled TXT fle for each pont whch corresponds to the vdeo fle Accordng to the eye tracker s specfcaton, the eye tracker records data about every /30 seconds, the data of each record ncludes seral number, tme, usage pattern, frame rate, eye observaton pont coordnates, pupl coordnates, and other data The eye tracker ncludes two modes of operaton, onlne and offlne Result shows that, there are slght fluctuatons n the tme of nterval between the two ponts recorded n dfferent modes, but t's roughly around /30 seconds Therefore, ths paper makes a reasonable smplfcaton as follows: defne the data record nterval of the Yarbus eye tracker as /30 seconds, show as t
Accordng to, human s eye movement produces two fxatons at least 0s, whch s far larger than the tme nterval between the two data recorded by eye tracker It can be deducted that the eye tracker not only records the fxaton ponts, but also the saccade ponts, the unconscous ponts and the nose ponts So t s needed to preprocess the eye tracker data, delete the error ponts, screen out the real ponts of regard, then analyss to mprove the accuracy of expermental data 33 Data Preprocessng 33 Analyss Process Ths artcle gves the followng defntons, defne the tme nterval between two fxaton ponts produced by human s eye movement as recorded by eye tracker as do ; defne the tme nterval between two data t ; defne human s pont of regard as ; defne the data pont recorded by eye tracker as Due to the mnmum value of s 0s, and t whch values as /30 seconds ddn t satsfy the defnton of the fxaton pont, t can be calculated: t 0/ 3 () t 30 If the data s recorded every /30 seconds, so 0s s evenly dvded nto three equal parts That s to say, the eye tracker record four data n 0s Because t s far less than,when the human eye movement s effectve fxaton, even though people are runnng durng the evacuaton,, the postons of ponts are almost unchanged n a very short perod of /30 seconds So, the four consecutve ponts should be very close to each other n 0s Therefore, the method to determne whether a pont s the pont s: determne whether the four consecutve ponts whch adjacent to each other n tme are also closely adjacent to each other n poston If the answer s yes, they are regarded as ;f on the contrary, they're regarded as the error ponts needed to be elmnated The method to determne whether four consecutve ponts are closely adjacent s as below: calculate whether the four ponts can be surrounded by an envelope crcle whose radus s settled The mathematcal explanaton s as shown n Fg 3: there are n ponts, the dstance between and + s n, r represents radus of the crcle When n r l, all the ponts can be surrounded n the crcle l, Fgure 3: The Envelope Crcle Dagram Includng n Ponts 33 Related Equaton Ths paper defned: the effectve length of vdeo s T, per frame tme of the vdeo s t the envelope crcle s x, y pont s, frame number s r n l, the coordnate of T n t, the coordnate of do pont s, radus of x, When the radus of envelope crcle meets the followng condton, t s consdered that four eye trackng ponts are ponts of regard T t r x x y y ( ) () If the four consecutve ponts are found, because the tme dfferences and poston dfferences among these four ponts are very small, the arthmetc mean value of ther coordnate s regarded as the coordnate of pont whch s shown as, x y x y x (3) y () POINT OF REGARD ANALYSIS In order to get the dstrbuton of the ponts, ths paper performs a seres of preprocessng towards the obtaned data For the determnaton of envelope radus sze, the radus was determned to be 6 based on repeated analyss and comparson The ponts dstrbuton of the two experments are shown n Fg and Fg 5 y 3
Fgure 6: Image Data By Grd Processng Fgure : Ponts of Regard Dstrbuton In Day-scenaro Experment Fgure 5: Ponts of Regard Dstrbuton In Nght-scenaro Experment The experments on day and nght were both evacuated from the 0th floor to the st floor, whose routes are consstent and the evacuaton tme are close The number of data ponts recorded by eye tracker s 5000 and 5500 respectvely, the number of ponts after data preprocessng are 53 and 0 respectvely Among them, when the evacuaton ndcator and pont are appearng n the same screen, the number s 6 and 83 respectvely The preprocessng results are shown n table Table : The Preprocessng Results pont s n the same frame do wth evacuaton ndcator Day 5000 53 63 Nght 5500 0 83 A further analyss was made on the ponts through preprocessng Fnd out the correspondng mage data for each pont, select the mage whch contans both pont and evacuaton ndcator The mage data are dstrbuted n dfferent areas of the buldng, ncludng corrdor and starcase There are only evacuaton sgns n the corrdor, and n the starwell there are both floor number sgns and evacuaton sgns Process every pcture by addng a 3 by 8 grd to make the scene as delcate as possble, as Fg 6 shows Accordng to the statstcs, the relatve poston dstrbuton between pont and evacuaton ndcator at dfferent locaton s shown n table Table : Relatve Poston Dstrbuton Between Scene Day Nght Pont And Evacuaton Indcator In Dfferent Area the relatve poston of do pont wth evacuaton ndcator evacuaton ndcator n corrdor floor number sgn n starcase do evacuaton ndcator n starcase top 3 0 0 mddle 5 5 0 bottom 0 5 top 0 0 mddle 9 5 bottom 7 5 It can be concluded that: the evacuaton sgns n the corrdor are usually placed near the low level of the ground, ths s because the smoke s heatng up when fre happens, and people are more lkely to bend down to see the sgns; however, the results show that, when people are evacuated n the corrdor, most of the people's ponts of regard are hgher than the evacuaton ndcator no matter day or nght Therefore, t s recommended to ncrease evacuaton ndcators at hgh place n the corrdor The mprovement s applcable to emergency evacuaton n smokeless condtons such as earthquake, terrorst attacks On the contrary, when people are evacuated n the starwell, ther ponts of regard are mostly lower than the floor number sgn as well as evacuaton ndcator, there s a close relatonshp wth person often focuses most of hs or her attenton on the feet when goes down the stars Therefore, t s suggested that the ndcaton poston of the starsshall be pasted on the floor of the starwell at the evacuaton floor and ground floor 5 CONCLUSIONS Ths paper adopts the method of emergency evacuaton experment, utlzes eye tracker as the expermental nstrument to obtan authentc and credble data; selects the fxaton pont based on the movng range of human s ponts of regard n the unt tme, and presents a new approach to the data of eye tracker The expermental result verfes the dstrbuton of the ponts of regard durng emergency evacuaton, studes the ratonalty of locaton dstrbuton of the exstng buldng evacuaton ndcators The
result shows that the locaton dstrbuton s a bt reasonable, there s stll space for mprovement, for example, t s recommended to ncrease the evacuaton ndcators at the hgh place of the corrdor for smokeless condtons; but nsde the starwell, floor number sgns and evacuaton nstructons should be moved down; and mark floor nformaton on the refuge floor and the ground floor However, the current method of data analyss s relatvely rough and qualtatve; the preprocessng method s lmted to the eye tracker brand used n the experment, t could only provde a tran of thought about data processng whch s lackng of unversalty The next step s to refne the data processng method, t s recommended to combne Mcrosoft's new HoloLens wth eye tracker, to buld a three-dmensonal model of the buldng, record each pont of regard n the three-dmensonal model ntutvely, and make a massve census Ths method s more accurate and quanttatve A HEADINGS IN APPENDICES The rules about herarchcal headngs dscussed above for the body of the artcle are derent n the appendces In the appendx envronment, the command secton s used to ndcate the start of each Appendx, wth alphabetc order desgnaton (e, the frst s A, the second B, etc) and a ttle (f you nclude one) So, f you need herarchcal structure wthn an Appendx, start wth subsecton as the hghest level Here s an outlne of the body of ths document n Appendx-approprate form: based on eye movement [J] Hunan transtaton technology, 0, 38 () : 53-55 [] hou Xny Applcaton of eye movement n landscape desgn and park gude map [D] hejang unversty, 05 [5] We hen Research on pedestran evacuaton mode based on behavoral feature analyss [D] South unversty of technology, 0 [6] hou Xueyan A prelmnary study on the gudng desgn of subway space supt [D] Shangha jaotong unversty, 009 [7] Lhua L, Nng Dng, Yapng Ma, Hu hang, Socal Relaton Network and Leader-Follower Behavor n Buldng Evacuaton 06, Fre Technology, 06 [8] Duchowsk A TEye trackng methodology:theory and practce[m]london:sprnger,003 [9] hao Xncan, uo Hongfu, Ren Yongjun Revew of eye movement and vson trackng [J] Computer engneerng and applcaton, 006, () : 8-0 [0] Cheng Shwe, Sun Lngyun Vsualzaton of eye movement data [J] Journal of computer aded desgn and graphc scence, 0, 6 (5) : 698-707 [] hang Xanyong Eye trackng pont postonng technology research [D] X 'an unversty of electronc scence and technology, 0 [] Andrew Duchowskc P Eye Trackng Methodology[M] Sprnger London, 007 [3] Sun Rushan, Chen Nongtan Eye movement analyss technology and ts applcaton n avaton feld [J] Journal of cvl avaton unversty, 009, 7 () : - [] Kassner M P, Patera W R PUPIL : constructng the space of vsual attenton[j] Massachusetts Insttute of Technology, 0 [5] Patera W, Patera W, Bullng A Pupl: an open source platform for pervasve eye trackng and moble gaze-based nteracton[c]// ACM Internatonal Jont Conference on Pervasve and Ubqutous Computng: Adjunct Publcaton ACM, 0:5-60 A Introducton A Experment Desgn A3 Related Work A3 A3 A33 A33 A33 Human s Eye Movement Eye Tracker s Orgnal Data Data Preprocessng Analyss Process Related Equaton A Pont of Regard Analyss A5 Conclusons A6 Acknowledgments A7 References ACKNOWLEDGMENTS The research s partally funded by the Natonal Natural Scence Foundaton of (Grant No9660) REFERENCES [] hang Sen, L Jngcheng, Xu Shousen, et al The development status of eye tracker and ts applcaton n sts psychology [J] Journal of captal sts college, 007, 9 () : 3-5 [] L Haqong, Qn Yaqn Applcaton of ophthalmoscope n the feld of road transtaton [J] Ergonomcs, 0, 8 () : 75-79 [3] hang Je Research on the dstrbuton characterstcs of drver's vewpont 5